Involvement of Machine Learning for Breast Cancer Image ...downloads.hindawi.com/journals/cmmm/2017/3781951.pdf · Involvement of Machine Learning for Breast Cancer Image Classification:
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Review ArticleInvolvement of Machine Learning for Breast CancerImage Classification A Survey
Abdullah-Al Nahid and Yinan Kong
School of Engineering Macquarie University Sydney NSW 2109 Australia
Correspondence should be addressed to Abdullah-Al Nahid abdullah-alnahidstudentsmqeduau
Received 29 August 2017 Accepted 26 October 2017 Published 31 December 2017
Academic Editor Po-Hsiang Tsui
Copyright copy 2017 Abdullah-Al Nahid and Yinan Kong This is an open access article distributed under the Creative CommonsAttribution License which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited
Breast cancer is one of the largest causes of womenrsquos death in the world today Advance engineering of natural image classificationtechniques and Artificial Intelligence methods has largely been used for the breast-image classification task The involvementof digital image classification allows the doctor and the physicians a second opinion and it saves the doctorsrsquo and physiciansrsquotime Despite the various publications on breast image classification very few review papers are available which provide a detaileddescription of breast cancer image classification techniques feature extraction and selection procedures classification measuringparameterizations and image classification findings We have put a special emphasis on the Convolutional Neural Network (CNN)method for breast image classification Along with the CNN method we have also described the involvement of the conventionalNeural Network (NN) Logic Based classifiers such as the Random Forest (RF) algorithm Support Vector Machines (SVM)Bayesianmethods and a few of the semisupervised and unsupervisedmethodswhich have been used for breast image classification
1 Introduction
The cell of the body maintains a cycle of regenerationprocesses The balanced growth and death rate of the cellsnormally maintain the natural working mechanism of thebody but this is not always the case Sometimes an abnormalsituation occurs where a few cells may start growing aber-rantly This abnormal growth of cells creates cancer whichcan start from any part of the body and be distributed to anyother part Different types of cancer can be formed in humanbody among them breast cancer creates a serious healthconcern Due to the anatomy of the human body womenare more vulnerable to breast cancer than men Among thedifferent reasons for breast cancer age family history breastdensity obesity and alcohol intake are reasons for breastcancer
Statistics reveal that in the recent past the situation hasbecome worse As a case study Figure 1 shows the breastcancer situation in Australia for the last 12 years This figurealso shows the number of new males and females to startsuffering frombreast cancer In 2007 the number of new cases
for breast cancer was 12775 while the expected number ofnew cancer patients in 2018 will be 18235 Statistics show thatin the last decade the number of new cancer disease patientsincreased every year at an alarming rate
Figure 2 shows the number of males and females facingdeath due to breast cancer It is predicted that in 2018 around3156 people will face death among them 3128 will be womenwhich is almost 9911 of the overall deaths due to breastcancer
Womenrsquos breasts are constructed by lobules ducts nip-ples and fatty tissues Milk is created in lobules and carriedtowards nipple by ducts Normally epithelial tumors growinside lobules as well as ducts and later form cancer inside thebreast [1] Once the cancer has started it also spreads to otherparts of the body Figure 3 shows the internal constructionfrom a breast image
Breast cancer tumors can be categorized into two broadscenarios
(i) Benign (Noncancerous) Benign cases are considered asnoncancerous that is non-life-threatening But on a few
HindawiComputational and Mathematical Methods in MedicineVolume 2017 Article ID 3781951 29 pageshttpsdoiorg10115520173781951
2 Computational and Mathematical Methods in Medicine
103
116
111
114
114
126
142
132
136
140
144
148
1267
2
1364
7
1380
6
1438
7
1456
9
1533
7
1590
2
1612
8
1656
6
1709
9
1758
6
1808
7
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
New
case
s
Year
MaleFemale
Figure 1 Number of new people facing cancer in Australia from2007 to 2018 [5]
25 14 29 23 22 24 29 30 26 27 28 28
2722
2746
2785
2837
2901
2823
2863
2814 30
01
3046
3087
3128
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
Year
MaleFemale
Num
ber o
f dea
ths
Figure 2 Number of people dying due to cancer in Australia from2007 to 2018 [5]
occasions it could turn into a cancer status An immunesystem known as ldquosacrdquo normally segregates benign tumorsfrom other cells and can be easily removed from the body
(ii) Malignant (Cancerous) Malignant cancer starts from anabnormal cell growth and might rapidly spread or invadenearby tissue Normally the nuclei of the malignant tissueare much bigger than in normal tissue which can be life-threatening in future stages
Cancer is always a life-threatening disease Proper treat-ment of cancer saves peoplersquos lives Identification of thenormal benign and malignant tissues is a very importantstep for further treatment of cancer For the identificationof benign and malignant conditions imaging of the targetedarea of the body helps the doctor and the physician infurther diagnosis With the advanced modern photographytechniques the image of the targeted part of the body can becaptured more reliably Based on the penetration of the skinand damage of the tissue medical photography techniquescan be classified into two groups
(i) Noninvasive (a) Ultrasound this photography techniqueuses similar techniques to SOund Navigation And Ranging(SONAR)which operates in the very-high-frequency domainand records the echos of that frequency invented by Karl
Theodore Dussik [2] An ultrasound imagemachine containsa Central Processing Unit (CPU) transducer a display unitand a few other peripheral devices This device is capable ofcapturing both 2D and 3D images Ultrasound techniquesdo not have any side-effects with some exceptions likeproduction of heat bubbles around the targeted tissue (b)X-ray X-rays utilize electromagnetic radiation invented byWilhelm Conrad Roentgen in 1895 The mammogram is aspecial kind of X-ray (low-dose) imaging technique whichis used to capture a detailed image of the breast [3] X-rayssometimes increase the hydrogen peroxide level of the bloodwhich may cause cell damage Sometimes X-rays may changethe base of DNA (c) Computer Aided Tomography (CAT)CAT or in short CT imaging is advanced engineering of X-ray imaging techniques where the X-ray images are takenat different angles The CT imaging technique was inventedin 1970 and has been mostly used for three-dimensionalimaging (d) Magnetic Resonance Imaging (MRI) MRI is anoninvasive imaging technique which produces a 3D imageof the body invented by Professor Sir Peter Marsfield andthis method utilizes both a magnetic field as well as radiowaves to capture the images [4] MRI techniques take longerto capture images which may create discomfort for the userExtra cautions need to be addressed to patients whomay haveimplanted extra metal
(ii) Invasive (a) Histopathological images (biopsy imaging)histopathology is the microscopic investigation of a tissueFor histopathological investigation a patient needs to gothrough a number of surgical steps The photographs takenfrom the histopathological tissue provide histopathologicalimages (see Figure 4)
2 Breast Image Classification
Various algorithms and investigation methods have beenused by researchers to investigate breast images fromdifferentperspectives depending on the demand of the disease thestatus of the disease and the quality of the images Amongthe different tasks for breast image classification machinelearning (ML) and the Artificial Intelligence (AI) are heavilyutilized A general breast image classifier consists of fourstages (see Figure 5)
(i) Selection of a breast database(ii) Feature extraction and selection(iii) Classifier model(iv) Performance measuring parameter(v) Classifier output
Figure 5 shows a very basic breast image classifier model
21 Available Breast Image Databases Doctors and physi-cians are heavily reliant on the ultrasoundMRI X-ray and soforth images to find the breast cancer present statusHoweverto ease the doctorsrsquo work some research groups are investi-gating how to use computers more reliably for breast cancerdiagnostics To make a reliable decision about the cancer
Computational and Mathematical Methods in Medicine 3
Figure 3 Anatomy of the female breast images (for the National Cancer Institute 2011 Terese Winslow US Government has certain rights)
(a) (b)
(c) (d)
Figure 4 (a b) showmammogram benign and malignant images (examples of noninvasive image) and (c d) show histopathological benignand malignant images (examples of invasive image)
4 Computational and Mathematical Methods in Medicine
Table 1 Available breast image database for biomedical investigation
Figure 5 A very basic breast image classification model
outcome researchers always base their investigation on somewell-established image database Various organizations haveintroduced sets of images databases which are available toresearchers for further investigation Table 1 gives a few of theavailable image databases with some specifications
The image formats of the different databases are differentFew of the images contained images in JPEG format and fewdatabases contained DICOM-format data Here the MIASDDSM and Inbreast databases containmammogram imagesAccording to the Springer (httpwwwspringercom)Elsevier (httpswwwelseviercom) and IEEE (httpwwwieeexploreieeeorg) web sites researchers have mostlyutilized the MIAS and DDSM databases for the breast imageclassification research The number of conference paperspublished for the DDSM and MIAS databases is 110 and 168respectively with 82 journal papers published on DDSMdatabases and 136 journal papers published using the MIASdatabase We have verified these statistics on both Scopus(httpswwwscopuscom) and the Web of Science database(httpwwwwebofknowledgecom) Figure 6 shows thenumber of published breast image classification papers basedon the MIAS and DDSM database from the years 2000 to2017
Histopathological images provide valuable informationand are being intensively investigated by doctors for find-ing the current situation of the patient The TCGA-BRCAand BreakHis databases contain histopathological imagesResearch has been performed in a few experiments on thisdatabase too Among these two databases BreakHis is themost recent histopathological image database containing a
4 4 3 2 47
16
68 8
2319
37
19
38
4541
17
1 0 04 4 4
7 8 96
1215 14
2123
2826
12
05
101520253035404550
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
Freq
uenc
y
Year
MIASDDSM
Figure 6 Number of papers published based on MIAS and DDSMdatabases
total of 7909 images which have been collected from 82patients [6] So far around twenty research papers have beenpublished based on this database
22 Feature Extraction and Selection An important stepof the image classification is extracting the features fromthe images In the conventional image classification taskfeatures are crafted locally using some specific rules andcriteria However the-state-of-the-art Convolutional NeuralNetwork (CNN) techniques generally extract the featuresglobally using kernels and these Global Features have beenused for image classification Among the local featurestexture detector and statistical are being accepted as impor-tant features for breast image classification Texture featuresactually represent the low-level feature information of animage which providesmore detailed information of an imagethat might be possible from histogram information aloneMore specifically texture features provide the structural anddimensional information of the color as well as the intensity
Computational and Mathematical Methods in Medicine 5
Table 2 Feature descriptor
Feature category Feature description
Texture
Haralick texture features [7]
(1) Angular Second Moment (ASM) (2) Contrast (3) correlation (4) Sum of Squares of Variances (SSoV) (5) Inverseof Difference (IoD) (6) Sum of Average (SoA) (7) Sum of Variances (SoV) (8) Sum of Entropy (SoE) (9) Entropy(10) Difference of Variance (DoV) (11) Difference of Entropy (DoE) (12) Gray-Level Concurrence Matrix (GLCM)Tamura features [8](1) Coarseness (2) Contrast (3) directionality (4) line-likeness (5) roughness (6) regularityGlobal texture descriptor(1) Fractal dimension (FD) (2) Coarseness (3) Entropy (4) Spatial Gray-Level Statistics (SGLS) (5) Circular MoranAutocorrelation Function (CMAF)
Detector
Single scale detector(1)Moravecrsquos Detector (MD) [9] (2)Harris Detector (HD) [10] (3) Smallest Univalue Segment Assimilating Nucleus(SUSAN) [11] (4) Features from Accelerated Segment Test (FAST) [12 13] (5)Hessian Blob Detector (HBD) [14 15]Multiscale detector [8](1) Laplacian of Gaussian (LoG) [9 16] (2) Difference of Gaussian (DoG) Contrast [17] (3)Harris Laplace (HL) (4)Hessian Laplace (HeL) (5) Gabor-Wavelet Detector (GWD) [18]
Figure 7 Classification of features for breast image classification
of the image Breast Imaging-Reporting and Data System(BI-RADS) is a mammography image assessment techniquecontaining 6 categories normally assigned by the radiologistFeature detector actually provides information whether theparticular feature is available in the image or not Structuralfeatures provide information about the features structure andorientation such as the area Convex Hull and centroid Thiskind of information gives more detailed information aboutthe features In a cancer image it can provide the area ofthe nucleus or the centroid of the mass Mean Medianand Standard Deviation always provide some importantinformation on the dataset and their distribution This kindof features has been categorized as statistical features Thetotal hierarchy of the image feature extraction is resented inFigure 7 Tables 2 and 3 further summarize the local featuresin detail
Features which are extracted for classification do notalways carry the same importance Some features may evencontribute to degrading the classifier performance Priori-tization of the feature set can reduce the classifier modelcomplexity and so it can reduce the computational timeFeature set selection and prioritization can be classified intothree broad categories
(i) Filter the filter method selects features without eval-uating any classifier algorithm
(ii) Wrapper the wrapper method selects the feature setbased on the evaluation performance of a particularclassifier
(iii) Embedded the embeddedmethod takes advantage ofthe filter andwrappermethods for classifier construc-tion
6 Computational and Mathematical Methods in Medicine
Table 3 Feature descriptor
Feature category Feature descriptionStatistical (1)Mean (2)Median (3) Standard Deviation (4) Skewness (5) Kurtosis (6) Range
Descriptor
(1) Scale Invariant Feature Transform (SIFT) [17 19] (2) Gradient Location-Orientation Histogram (GLOH) [20] (3)Speeded-Up Robust Features Descriptor (SURF) [21ndash23] (4) Local Binary Pattern (LBP) [24ndash27] (5) Binary RobustIndependent Elementary Features (BRIEF) [28] (6)Weber Local Descriptor (WLD) [29 30] (7) Back Ground LocalBinary Pattern (BGLBP) [31] (8) Center-Symmetric Local Binary Pattern (CS-LBP) [32] (9) Second-OrderCenter-Symmetric Local Derivative Pattern (CS-LBP) [33] (10) Center-Symmetric Scale Invariant Local TernaryPatterns (CS-SILTP) [34] (11) Extended LBP or Circular LBP (E-LBP) [35] (12)Opponent Color Local Binary Pattern(OC-LBP) [36] (13) Original LBP(O-LBP) [25] (14) Spatial Extended Center-Symmetric Local Binary Pattern(SCS-LBP) [37] (15) Scale Invariant Local Ternary Pattern (SI-LTP) [38] (16) Variance-Based LBP (VAR-LBP) [24](17) eXtended Center-Symmetric Local Binary Pattern (XCS-LBP) (18) Average Local Binary Pattern (ALBP) (19)Block Based Local Binary Pattern (BBLBP) [39]
Figure 8 shows a generalized feature selection methodwhere we have further classified the filter method intoFisher Score Mutual Information Relief and chi squaremethods The embedded method has been classified intoBridge Regularization Lasso and Adaptive Lasso methodswhile the wrapper method has been classified to recursivefeature selection and sequential feature selection method
23 Classifier Model Based on the learning point of viewbreast image classification techniques can be categorized intothe following three classes [41]
These three classes can be split into Deep Neural Network(DNN) and conventional classifier (without DNN) and tosome further classes as in Table 4
24 Performance Measuring Parameter A Confusion Matrixis a two-dimensional table which is used to a give a visual
True
clas
s
Hypothesized class
True positive (A) False negative (B)
False positive (C) True negative (D)
Figure 9 Confusion Matrix
perception of classification experiments [54] The (119894 119895)thposition of the confusion table indicates the number of timesthat the 119894th object is classified as the 119895th object The diagonalof this matrix indicates the number of times the objects arecorrectly classified Figure 9 shows a graphical representationof a Confusion Matrix for the binary classification case
Computational and Mathematical Methods in Medicine 7
Table 4 A simplified hierarchy of classification
Learning technique Algorithm
Supervised
Conventional
(a) Logic based
(1) ID3 (2) C45 (3) bagging(4) random trees (5) Random Forest(6) boosting (7) advanced boosting(8) Extreme Boosting (XGBoosting)
(a) Self-training(b) Graph Based(c) S3V3(d) Multiview(e) Generative model
Among the different classification performance proper-ties this matrix will provide following parameters
(i) Recall is defined as Recall = TP(TP + FN)(ii) Precision is defined as Precision = TP(TP + FP)(iii) Specificity is defined as Specificity = TN(TN + FP)(iv) Accuracy is defined as ACC = (TP+TN)(TP+TN+
FP + FN)(v) F-1 score is defined as 1198651 = (2 times Recall)(2 times Recall +
FP + FN)(vi) Matthew Correlation Coefficient (MCC) MCC is a
performance parameter of a binary classifier in therange minus1 to +1 If the MCC values trend moretowards +1 the classifier gives a more accurate classi-fier and the opposite condition will occur if the valueof theMCC trend towards theminus1MCCcanbe definedas
MCC
= TP times TN minus FP times FNradic(TP + FP) (TP + FN) (TN + FP) (TN + FP) (1)
3 Performance of Different Classifier Modelon Breast Images Dataset
Based on Supervised Semisupervised and Unsupervisedmethods different research groups have been performedclassification operation on different image database In thissection we have summarized few of the works of breast imageclassification
31 Performance Based on Supervised Learning In super-vised learning a general hypothesis is established based onexternally supplied instances to produce future predictionFor the supervised classification task features are extractedor automatically crafted from the available dataset and eachsample is mapped to a dedicated class With the help of thefeatures and their levels a hypothesis is created Based on thehypothesis unknown data are classified [55]
Figure 10 represents an overall supervised classifier archi-tecture In general the whole dataset is split into trainingand testing parts To validate the data some time dataare also split into a validation part as well After the datasplitting themost important part is to find out the appropriatefeatures to classify the data with the utmost AccuracyFinding the features can be classified into two categorieslocally and globally crafted Locally crafted means that thismethod requires a hand-held exercise to find out the featureswhereas globally craftedmeans that a kernelmethod has beenintroduced for the feature extraction Handcrafted featurescan be prioritized whereas Global Feature selection does nothave this luxury
311 Conventional Neural Network The Neural Network(NN) concept comes from the working principle of thehuman brain A biological neuron consists of the followingfour parts
8 Computational and Mathematical Methods in Medicine
Classifier model
Imagedatabase
Traintestdata splitting Locally
craftedGloballycrafted
Hand crafting
Kernel basedcrafting
Featureprioritization
Conventionalclassifier
DNNclassifier
Evaluationmatrix
Classifieddata
Feature collection
Ensemble learning
Figure 10 A generalized supervised classifier model
Nucleus
Axon
Cell body
Dendrites
Figure 11 A model of a biological neuron
Dendrites collect signals and axons carry the signal to thenext dendrite after processing by the cell body as shown inFigure 11 Using the neuronworking principle the perceptronmodel was proposed by Rosenblatt in 1957 [56] A single-layer perceptron linearly combines the input signal and givesa decision based on a threshold function Based on theworking principle and with some advanced mechanism andengineering NNmethods have established a strong footprintin many problem-solving issues Figure 12 shows the basicworking principle of NN techniques
In the NN model the input data X = 1199090 1199091 119909119873 isfirst multiplied by the weight dataW = 1199080 1199081 119908119873 andthen the output is calculated using
Y = g (sum) wheresum = W sdot X (2)
Function g is known as the activation function Thisfunction can be any threshold value or Sigmoid or hyperbolicand so forth In the early stages feed-forwardNeuralNetworktechniques were introduced [57] lately the backpropagationmethod has been invented to utilize the error information toimprove the system performance [58 59]
The history of breast image classification by NN is a longone To the best of my knowledge a lot of the pioneer work
yg
x0
x1
xNminus1
xN
w0
w1
wNminus1
wN
Figure 12Working principle of a simpleNeuralNetwork technique
was performed by Dawson et al in 1991 [60] Since then NNhas been utilized as one of the strong tools for breast imageclassification We have summarized some of the work relatedto NN and breast image classification in Tables 5 6 and 7
312 Deep Neural Network Deep Neural Network (DNN) isa state-of-the-art concept where conventional NN techniqueshave been utilized with advanced engineering It is foundthat conventional NNs have difficulties in solving complexproblems whereas DNNs solve them with utmost PrecisionHowever DNNs suffer from more time and computationalcomplexity than the conventional NN
Convolutional Neural Network A CNN model is the combi-nation of a few intermediate mathematical structures Thisintermediatemathematical structure creates or helps to createdifferent layers
(i) Convolutional Layer Among all the other layers theconvolutional layer is considered as the most important partfor a CNN model and can be considered as the backbone of
Computational and Mathematical Methods in Medicine 9
Table 5 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Rajakeerthana et al [42] (1) GLCM GLDM SRDMNGLCM GLRM Mammogram 322 (1)The classifier achieved 9920
Accuracy
Lessa and Marengoni [43](1)Mean Median StandardDeviation Skewness KurtosisEntropy Range
Wan et al [44] (1) ALBP (2) BBLBP OCM 46(1) Achieved Sensitivity and Specificityare 100 and 8520 respectively(2) ROC value obtained 0959
Chen et al [40] (1) 19 BI-RADS features havebeen used Ultrasound 238
(1) Chi squared method has beenutilized for the feature selection(2) Achieved Accuracy Sensitivity andSpecificity are 9610 9670 and9570 respectively
de Lima et al [45] (1) Total 416 features have beenused Mammogram 355
(1)Multiresolution wavelet and Zernikemoment have been utilized for thefeature extraction
Abirami et al [46](1) 12 statistical measures such asMean Median and Max havebeen utilized as the features
Mammogram 322
(1)Wavelet transform has been utilizedfor the feature extraction(2)The achieved Accuracy Sensitivityand Specificity are 9550 9500 and9600 respectively
El Atlas et al [47] (1) 13 morphological featureshave been utilized Mammogram 410
(1) Firstly the edge information hasbeen utilized for the mass segmentationand then the morphological featureswere extracted(2) Achieved best Accuracy is 975
Table 6 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Alharbi et al [48] (1) 49 features havebeen utilized Mammogram 1100
(1) Five feature selection methods Fisher scoreMinimum Redundancy-Maximum Relevance Relief-fSequential Forward Feature Selection and GeneticAlgorithm have been used(2) Achieved Accuracy Sensitivity and specificity are9420 9836 and 9927 respectively
Peng et al [49](1)Haralick andTamura features havebeen utilized
Mammogram 322
(1) Feature reduction has been performed byRough-Set theory and selected 5 prioritized features(2)The best Accuracy Sensitivity and Specificityachieved were 9600 9860 and 8930
Jalalian et al [50] (1) GLCM Mammogram(1)The obtained classifier Accuracy Sensitivity andSpecificity are 9520 9240 and 9800respectively(2) Compactness
Li et al [51](1) Four featurevectors have beencalculated
Mammogram 322
(1) 2D contour of breast mass in mammography hasbeen converted into 1D signature(2) NN techniques achieved Accuracy is 9960 whenRMS slope is utilized
Chen et al [52] (1) Autocorrelationfeatures Ultrasound 242 (1)The overall achieved Accuracy Sensitivity and
Specificity are 9500 9800 and 93 respectively
Chen et al [53] (1) Autocorrelationfeatures Ultrasound 1020 (1)The obtained ROC area is 09840 plusmn 00072
10 Computational and Mathematical Methods in Medicine
Table 7 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Chen et al [61]
(1) Variance Contrast of WaveletCoefficient Ultrasound 242 (1)The achieved ROC curve 09396 plusmn 00183(2) Autocorrelation of WaveletCoefficient
Silva et al [62](1) 22 different morphologicalfeatures such as convexity andlobulation have been utilized
Ultrasound mdash (1)The best obtained Accuracy and ROCcurve are 9698 and 098 respectively
Saritas [63](1) Age of patient (2)massshape (3)mass border (4)Massdensity (5) BIRADS Mammogram mdash
(1) Disease prediction rate is 905(2) Neural Network utilized 5 neurons ininput layers and one hidden layer
Lopez-Melendez etal [64]
(1) Area perimeter etc havebeen utilized Mammogram 322 (1)The achieved Sensitivity and Specificity
are 9629 and 9900 respectively
themodel A kernel of size119898times119899 is scanned through the inputdata for the convolutional operation which ensures the localconnectivity and weight sharing property
(ii) Stride and Padding In the convolutional operation afilter scans through the input matrices In each step howmuch position a kernel filter moves through the matrixis known as the stride By default stride keeps to 1 Withinappropriate selection of the stride the model can lose theborder information To overcome this issue themodel utilizesextra rows and columns at the end of the matrices and theseadded rows and columns contain all 0s This adding of extrarows and columns which contain only zero value is known aszero padding
(iii) Nonlinear Operation The output of each of the kerneloperations is passed through a rectifier function such as Rec-tified Linear Unit (ReLU) Leaky-ReLU TanH and SigmoidThe Sigmoid function can be defined as
120590 (119909) = 1(1 + expminus119909) (3)
and the tanh function can be defined as
tanh (119909) = (exp119909 minus expminus119909)(exp119909 + expminus119909) (4)
However the most effective rectifier is ReLU The ReLUmethod converts all the information into zero if it is less thanor equal to zero and passes all the other data as is shown inFigure 13
120590 (119909) = max (0 119909) (5)
Another important nonlinear function is Leaky-RelU
where 120572 is predetermined parameter which can be varied togive a better model
minus3 minus2 minus1 0 1 2 3
1
2
3
InputO
utpu
t
Figure 13 ReLU Operation
(iv) Subsampling Subsampling is the procedure of reducingthe dimensionality of each of the feature maps of a particularlayer this operation is also known as a pooling operationActually it reduces the amount of feature information fromthe overall data By doing so it reduces the overall computa-tional complexity of themodel To do this 119904times119904 patch units areutilized The two most popular pooling methods are
(a) Max-Pooling
(b) Average Pooling
In Max-Pooling only the maximum values within a partic-ular kernel size are selected for further calculation Consideran example of a 16 times 16 image as shown in Figure 14 A 2 by2 kernel is applied to the whole image 4 blocks in total andproduces a 4 times 4 output image For each block of four valueswe have selected the maximum For instance from blocksone two three and four maximum values 4 40 13 and 8are selected respectively as they are the maximum in thatblock For the Average Pooling operation each kernel givesthe output as average
(v) Dropout Regularization of the weight can reduce theoutfitting problem Randomly removing some neurons can
Computational and Mathematical Methods in Medicine 11
Figure 15 Work-flow of a Convolutional Neural Network
regularize the overfilling problem The technique of ran-domly removing neurons from the network is known asdropout
(vi) Soft-Max Layer This layer contains normalized expo-nential functions to calculate the loss function for the dataclassification
Figure 15 shows a generalized CNN model for the imageclassificationAll the neurons of themost immediate layer of afully connected layer are completely connected with the fullyconnected layer like a conventional Neural Network Let119891119897minus1119895represent the 119895th feature map at the layer 119897minus1The 119895th featuremap at the layer 119897 can be represented as
where119873119897minus119897 represents the number of featuremaps at the 119897minus1thlayer 119896119894119895 represents the kernel function and 119887119897119895 represents thebias at 119897 where 120590 performs a nonlinear function operationThe layer before the Soft-Max Layer can be represented as
Let 119901 = 1 represent Benign class and 119901 = 2 represent theMalignant class The cross-entropy loss of the above functioncan be calculated as
119871119901 = minus ln (119910119901) (10)
Whichever group experiences a large loss value themodel will consider the other group as predicted class
A difficult part of working on DNN is that it requiresa specialized software package for the data analysis Fewresearch groups have been working on how effectively datacan be analyzed by DNN from different perspectives and thedemand Table 8 summarizes some of the software which isavailable for DNN analysis
The history of the CNN and its use for biomedical imageanalysis is a long one Fukushima first introduced a CNNnamed ldquonecognitronrdquo which has the ability to recognizestimulus patterns with a few shifting variances [113] Tothe best of our knowledge Wu et al first classified a setof mammogram images into malignant and benign classesusing a CNN model [78] In their proposed model they onlyutilized one hidden layer After that in 1996 Sahiner et alutilized CNNmodel to classify mass and normal breast tissueand achieved ROC scores of 087 [79] In 2002 Lo et alutilized aMultiple Circular Path CNN (MCPCNN) for tumoridentification from mammogram images and obtained ROCscores of around 089 After an absence of investigation ofthe CNN model this model regained its momentum afterthe work of Krizhevsky et al [114] Their proposed model isknown as AlexNet After this work a revolutionary change
12 Computational and Mathematical Methods in Medicine
Table 8 Available software for deep learning analysis
Software Interface and backend Provider
Caffe [65 66] Python MATLAB C++ Berkeley Vision and Learning CentreUniversity of California Berkeley
Torch [67] C LuaJIT
MatConvNet [68 69] MATLAB C Visual Geometry Group Department ofEngineering University of Oxford
Theano [70 71] Python Montreal Institute for Learning AlgorithmsUniversity of Montreal
TensorFlows [72] C++ Python GoogleCNTK [73] C++ MicrosoftKeras [74] Theano Tensor Flow MITdl4j [75] Java Skymind Engineering
DeeBNET [76 77] MATLAB Information Technology DepartmentAmirkabir University of Technology
has been achieved in the image classification and analysisfield As an advanced engineering of the AlexNet the papertitled ldquoGoing Deeper with Convolutionsrdquo by Szegedy [115]introduced the GoogleNet model This model contains amuch deeper network than AlexNet Sequentially ResNet[116] Inception [117] Inception-v4 Inception-ResNet [118]and a few other models have recently been introduced
Later directly or with some advanced modificationthese DNN models have been adapted for biomedical imageanalysis In 2015 Fonseca et al [81] classified breast densityusing CNN techniques CNN requires a sufficient amountof data to train the system It is always very difficult tofind a sufficient amount of medical data for training a CNNmodel A pretrained CNN model with some fine tuning canbe used rather than create a model from scratch [119] Theauthors of [119] did not perform their experiments on a breastcancer image dataset however they have performed theirexperiments on three different medical datasets with layer-wise training and claimed that ldquoretrained CNN along withadequate training can provide better or at least the sameamount of performancerdquo
The Deep Belief Network (DBN) is another branch of theDeep Neural Network which mainly consists of RestrictedBoltzmann Machine (RBM) techniques The DBN methodwas first utilized for supervised image classification by Liu etal [120] After that Abdel-Zaher and Eldeib utilized the DBNmethod for breast image classification [121] This field is stillnot fully explored for breast image classification yet Zhanget al utilized both RBM and Point-Wise Gated RBM (PRBM)for shear-wave electrography image classification where thedataset contains 227 images [97]Their achieved classificationAccuracy Sensitivity and Specificity are 9340 8860 and9710 respectively Tables 9 10 and 11 have summarized themost recent work for breast image classification along withsome pioneer work on CNN
313 Logic Based Algorithm A Logic Based algorithm isa very popular and effective classification method whichfollows the tree structure principle and logical argument asshown in Figure 16 This algorithm classifies instances based
on the featurersquos values Along with other criteria a decision-tree based algorithm contains the following features
(i) Root node a root node contains no incoming nodeand it may or may not contain any outgoing edge
(ii) Splitting splitting is the process of subdividing a set ofcases into a particular group Normally the followingcriteria are maintained for the splitting
(a) information gain(b) Gini index(c) chi squared
(iii) Decision node(iv) Leafterminal node this kind of node has exactly one
incoming edge and no outgoing edgeThe tree alwaysterminates here with a decision
(v) Pruning pruning is a process of removing subtreesfrom the tree Pruning performs to reduce the over-fitting problem Two kinds of pruning techniques areavailable
(a) prepruning(b) postpruning
Among all the tree based algorithms IterativeDichotomiser 3 (ID3) can be considered as a pioneerproposed by Quinlan [149] The problem of the ID3algorithm is to find the optimal solution which is very muchprone towards overfitting To overcome the limitation of theID3 algorithm the C45 algorithm has been introduced byQuinlan [150] where a pruning method has been introducedto control the overfitting problem Pritom et al [151] classifiedthe Wisconsin breast dataset where they utilized 35 featuresThey have obtained 7630 Accuracy 7510 False PositiveRate and ROC score 0745 when they ranked the featuresWithout ranking the features they obtained 7370Accuracy5070 False Positive Rate and ROC score value 5280 Asriet al [152] utilized the C45 algorithm for the Wisconsin
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
2 Computational and Mathematical Methods in Medicine
103
116
111
114
114
126
142
132
136
140
144
148
1267
2
1364
7
1380
6
1438
7
1456
9
1533
7
1590
2
1612
8
1656
6
1709
9
1758
6
1808
7
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
New
case
s
Year
MaleFemale
Figure 1 Number of new people facing cancer in Australia from2007 to 2018 [5]
25 14 29 23 22 24 29 30 26 27 28 28
2722
2746
2785
2837
2901
2823
2863
2814 30
01
3046
3087
3128
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
Year
MaleFemale
Num
ber o
f dea
ths
Figure 2 Number of people dying due to cancer in Australia from2007 to 2018 [5]
occasions it could turn into a cancer status An immunesystem known as ldquosacrdquo normally segregates benign tumorsfrom other cells and can be easily removed from the body
(ii) Malignant (Cancerous) Malignant cancer starts from anabnormal cell growth and might rapidly spread or invadenearby tissue Normally the nuclei of the malignant tissueare much bigger than in normal tissue which can be life-threatening in future stages
Cancer is always a life-threatening disease Proper treat-ment of cancer saves peoplersquos lives Identification of thenormal benign and malignant tissues is a very importantstep for further treatment of cancer For the identificationof benign and malignant conditions imaging of the targetedarea of the body helps the doctor and the physician infurther diagnosis With the advanced modern photographytechniques the image of the targeted part of the body can becaptured more reliably Based on the penetration of the skinand damage of the tissue medical photography techniquescan be classified into two groups
(i) Noninvasive (a) Ultrasound this photography techniqueuses similar techniques to SOund Navigation And Ranging(SONAR)which operates in the very-high-frequency domainand records the echos of that frequency invented by Karl
Theodore Dussik [2] An ultrasound imagemachine containsa Central Processing Unit (CPU) transducer a display unitand a few other peripheral devices This device is capable ofcapturing both 2D and 3D images Ultrasound techniquesdo not have any side-effects with some exceptions likeproduction of heat bubbles around the targeted tissue (b)X-ray X-rays utilize electromagnetic radiation invented byWilhelm Conrad Roentgen in 1895 The mammogram is aspecial kind of X-ray (low-dose) imaging technique whichis used to capture a detailed image of the breast [3] X-rayssometimes increase the hydrogen peroxide level of the bloodwhich may cause cell damage Sometimes X-rays may changethe base of DNA (c) Computer Aided Tomography (CAT)CAT or in short CT imaging is advanced engineering of X-ray imaging techniques where the X-ray images are takenat different angles The CT imaging technique was inventedin 1970 and has been mostly used for three-dimensionalimaging (d) Magnetic Resonance Imaging (MRI) MRI is anoninvasive imaging technique which produces a 3D imageof the body invented by Professor Sir Peter Marsfield andthis method utilizes both a magnetic field as well as radiowaves to capture the images [4] MRI techniques take longerto capture images which may create discomfort for the userExtra cautions need to be addressed to patients whomay haveimplanted extra metal
(ii) Invasive (a) Histopathological images (biopsy imaging)histopathology is the microscopic investigation of a tissueFor histopathological investigation a patient needs to gothrough a number of surgical steps The photographs takenfrom the histopathological tissue provide histopathologicalimages (see Figure 4)
2 Breast Image Classification
Various algorithms and investigation methods have beenused by researchers to investigate breast images fromdifferentperspectives depending on the demand of the disease thestatus of the disease and the quality of the images Amongthe different tasks for breast image classification machinelearning (ML) and the Artificial Intelligence (AI) are heavilyutilized A general breast image classifier consists of fourstages (see Figure 5)
(i) Selection of a breast database(ii) Feature extraction and selection(iii) Classifier model(iv) Performance measuring parameter(v) Classifier output
Figure 5 shows a very basic breast image classifier model
21 Available Breast Image Databases Doctors and physi-cians are heavily reliant on the ultrasoundMRI X-ray and soforth images to find the breast cancer present statusHoweverto ease the doctorsrsquo work some research groups are investi-gating how to use computers more reliably for breast cancerdiagnostics To make a reliable decision about the cancer
Computational and Mathematical Methods in Medicine 3
Figure 3 Anatomy of the female breast images (for the National Cancer Institute 2011 Terese Winslow US Government has certain rights)
(a) (b)
(c) (d)
Figure 4 (a b) showmammogram benign and malignant images (examples of noninvasive image) and (c d) show histopathological benignand malignant images (examples of invasive image)
4 Computational and Mathematical Methods in Medicine
Table 1 Available breast image database for biomedical investigation
Figure 5 A very basic breast image classification model
outcome researchers always base their investigation on somewell-established image database Various organizations haveintroduced sets of images databases which are available toresearchers for further investigation Table 1 gives a few of theavailable image databases with some specifications
The image formats of the different databases are differentFew of the images contained images in JPEG format and fewdatabases contained DICOM-format data Here the MIASDDSM and Inbreast databases containmammogram imagesAccording to the Springer (httpwwwspringercom)Elsevier (httpswwwelseviercom) and IEEE (httpwwwieeexploreieeeorg) web sites researchers have mostlyutilized the MIAS and DDSM databases for the breast imageclassification research The number of conference paperspublished for the DDSM and MIAS databases is 110 and 168respectively with 82 journal papers published on DDSMdatabases and 136 journal papers published using the MIASdatabase We have verified these statistics on both Scopus(httpswwwscopuscom) and the Web of Science database(httpwwwwebofknowledgecom) Figure 6 shows thenumber of published breast image classification papers basedon the MIAS and DDSM database from the years 2000 to2017
Histopathological images provide valuable informationand are being intensively investigated by doctors for find-ing the current situation of the patient The TCGA-BRCAand BreakHis databases contain histopathological imagesResearch has been performed in a few experiments on thisdatabase too Among these two databases BreakHis is themost recent histopathological image database containing a
4 4 3 2 47
16
68 8
2319
37
19
38
4541
17
1 0 04 4 4
7 8 96
1215 14
2123
2826
12
05
101520253035404550
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
Freq
uenc
y
Year
MIASDDSM
Figure 6 Number of papers published based on MIAS and DDSMdatabases
total of 7909 images which have been collected from 82patients [6] So far around twenty research papers have beenpublished based on this database
22 Feature Extraction and Selection An important stepof the image classification is extracting the features fromthe images In the conventional image classification taskfeatures are crafted locally using some specific rules andcriteria However the-state-of-the-art Convolutional NeuralNetwork (CNN) techniques generally extract the featuresglobally using kernels and these Global Features have beenused for image classification Among the local featurestexture detector and statistical are being accepted as impor-tant features for breast image classification Texture featuresactually represent the low-level feature information of animage which providesmore detailed information of an imagethat might be possible from histogram information aloneMore specifically texture features provide the structural anddimensional information of the color as well as the intensity
Computational and Mathematical Methods in Medicine 5
Table 2 Feature descriptor
Feature category Feature description
Texture
Haralick texture features [7]
(1) Angular Second Moment (ASM) (2) Contrast (3) correlation (4) Sum of Squares of Variances (SSoV) (5) Inverseof Difference (IoD) (6) Sum of Average (SoA) (7) Sum of Variances (SoV) (8) Sum of Entropy (SoE) (9) Entropy(10) Difference of Variance (DoV) (11) Difference of Entropy (DoE) (12) Gray-Level Concurrence Matrix (GLCM)Tamura features [8](1) Coarseness (2) Contrast (3) directionality (4) line-likeness (5) roughness (6) regularityGlobal texture descriptor(1) Fractal dimension (FD) (2) Coarseness (3) Entropy (4) Spatial Gray-Level Statistics (SGLS) (5) Circular MoranAutocorrelation Function (CMAF)
Detector
Single scale detector(1)Moravecrsquos Detector (MD) [9] (2)Harris Detector (HD) [10] (3) Smallest Univalue Segment Assimilating Nucleus(SUSAN) [11] (4) Features from Accelerated Segment Test (FAST) [12 13] (5)Hessian Blob Detector (HBD) [14 15]Multiscale detector [8](1) Laplacian of Gaussian (LoG) [9 16] (2) Difference of Gaussian (DoG) Contrast [17] (3)Harris Laplace (HL) (4)Hessian Laplace (HeL) (5) Gabor-Wavelet Detector (GWD) [18]
Figure 7 Classification of features for breast image classification
of the image Breast Imaging-Reporting and Data System(BI-RADS) is a mammography image assessment techniquecontaining 6 categories normally assigned by the radiologistFeature detector actually provides information whether theparticular feature is available in the image or not Structuralfeatures provide information about the features structure andorientation such as the area Convex Hull and centroid Thiskind of information gives more detailed information aboutthe features In a cancer image it can provide the area ofthe nucleus or the centroid of the mass Mean Medianand Standard Deviation always provide some importantinformation on the dataset and their distribution This kindof features has been categorized as statistical features Thetotal hierarchy of the image feature extraction is resented inFigure 7 Tables 2 and 3 further summarize the local featuresin detail
Features which are extracted for classification do notalways carry the same importance Some features may evencontribute to degrading the classifier performance Priori-tization of the feature set can reduce the classifier modelcomplexity and so it can reduce the computational timeFeature set selection and prioritization can be classified intothree broad categories
(i) Filter the filter method selects features without eval-uating any classifier algorithm
(ii) Wrapper the wrapper method selects the feature setbased on the evaluation performance of a particularclassifier
(iii) Embedded the embeddedmethod takes advantage ofthe filter andwrappermethods for classifier construc-tion
6 Computational and Mathematical Methods in Medicine
Table 3 Feature descriptor
Feature category Feature descriptionStatistical (1)Mean (2)Median (3) Standard Deviation (4) Skewness (5) Kurtosis (6) Range
Descriptor
(1) Scale Invariant Feature Transform (SIFT) [17 19] (2) Gradient Location-Orientation Histogram (GLOH) [20] (3)Speeded-Up Robust Features Descriptor (SURF) [21ndash23] (4) Local Binary Pattern (LBP) [24ndash27] (5) Binary RobustIndependent Elementary Features (BRIEF) [28] (6)Weber Local Descriptor (WLD) [29 30] (7) Back Ground LocalBinary Pattern (BGLBP) [31] (8) Center-Symmetric Local Binary Pattern (CS-LBP) [32] (9) Second-OrderCenter-Symmetric Local Derivative Pattern (CS-LBP) [33] (10) Center-Symmetric Scale Invariant Local TernaryPatterns (CS-SILTP) [34] (11) Extended LBP or Circular LBP (E-LBP) [35] (12)Opponent Color Local Binary Pattern(OC-LBP) [36] (13) Original LBP(O-LBP) [25] (14) Spatial Extended Center-Symmetric Local Binary Pattern(SCS-LBP) [37] (15) Scale Invariant Local Ternary Pattern (SI-LTP) [38] (16) Variance-Based LBP (VAR-LBP) [24](17) eXtended Center-Symmetric Local Binary Pattern (XCS-LBP) (18) Average Local Binary Pattern (ALBP) (19)Block Based Local Binary Pattern (BBLBP) [39]
Figure 8 shows a generalized feature selection methodwhere we have further classified the filter method intoFisher Score Mutual Information Relief and chi squaremethods The embedded method has been classified intoBridge Regularization Lasso and Adaptive Lasso methodswhile the wrapper method has been classified to recursivefeature selection and sequential feature selection method
23 Classifier Model Based on the learning point of viewbreast image classification techniques can be categorized intothe following three classes [41]
These three classes can be split into Deep Neural Network(DNN) and conventional classifier (without DNN) and tosome further classes as in Table 4
24 Performance Measuring Parameter A Confusion Matrixis a two-dimensional table which is used to a give a visual
True
clas
s
Hypothesized class
True positive (A) False negative (B)
False positive (C) True negative (D)
Figure 9 Confusion Matrix
perception of classification experiments [54] The (119894 119895)thposition of the confusion table indicates the number of timesthat the 119894th object is classified as the 119895th object The diagonalof this matrix indicates the number of times the objects arecorrectly classified Figure 9 shows a graphical representationof a Confusion Matrix for the binary classification case
Computational and Mathematical Methods in Medicine 7
Table 4 A simplified hierarchy of classification
Learning technique Algorithm
Supervised
Conventional
(a) Logic based
(1) ID3 (2) C45 (3) bagging(4) random trees (5) Random Forest(6) boosting (7) advanced boosting(8) Extreme Boosting (XGBoosting)
(a) Self-training(b) Graph Based(c) S3V3(d) Multiview(e) Generative model
Among the different classification performance proper-ties this matrix will provide following parameters
(i) Recall is defined as Recall = TP(TP + FN)(ii) Precision is defined as Precision = TP(TP + FP)(iii) Specificity is defined as Specificity = TN(TN + FP)(iv) Accuracy is defined as ACC = (TP+TN)(TP+TN+
FP + FN)(v) F-1 score is defined as 1198651 = (2 times Recall)(2 times Recall +
FP + FN)(vi) Matthew Correlation Coefficient (MCC) MCC is a
performance parameter of a binary classifier in therange minus1 to +1 If the MCC values trend moretowards +1 the classifier gives a more accurate classi-fier and the opposite condition will occur if the valueof theMCC trend towards theminus1MCCcanbe definedas
MCC
= TP times TN minus FP times FNradic(TP + FP) (TP + FN) (TN + FP) (TN + FP) (1)
3 Performance of Different Classifier Modelon Breast Images Dataset
Based on Supervised Semisupervised and Unsupervisedmethods different research groups have been performedclassification operation on different image database In thissection we have summarized few of the works of breast imageclassification
31 Performance Based on Supervised Learning In super-vised learning a general hypothesis is established based onexternally supplied instances to produce future predictionFor the supervised classification task features are extractedor automatically crafted from the available dataset and eachsample is mapped to a dedicated class With the help of thefeatures and their levels a hypothesis is created Based on thehypothesis unknown data are classified [55]
Figure 10 represents an overall supervised classifier archi-tecture In general the whole dataset is split into trainingand testing parts To validate the data some time dataare also split into a validation part as well After the datasplitting themost important part is to find out the appropriatefeatures to classify the data with the utmost AccuracyFinding the features can be classified into two categorieslocally and globally crafted Locally crafted means that thismethod requires a hand-held exercise to find out the featureswhereas globally craftedmeans that a kernelmethod has beenintroduced for the feature extraction Handcrafted featurescan be prioritized whereas Global Feature selection does nothave this luxury
311 Conventional Neural Network The Neural Network(NN) concept comes from the working principle of thehuman brain A biological neuron consists of the followingfour parts
8 Computational and Mathematical Methods in Medicine
Classifier model
Imagedatabase
Traintestdata splitting Locally
craftedGloballycrafted
Hand crafting
Kernel basedcrafting
Featureprioritization
Conventionalclassifier
DNNclassifier
Evaluationmatrix
Classifieddata
Feature collection
Ensemble learning
Figure 10 A generalized supervised classifier model
Nucleus
Axon
Cell body
Dendrites
Figure 11 A model of a biological neuron
Dendrites collect signals and axons carry the signal to thenext dendrite after processing by the cell body as shown inFigure 11 Using the neuronworking principle the perceptronmodel was proposed by Rosenblatt in 1957 [56] A single-layer perceptron linearly combines the input signal and givesa decision based on a threshold function Based on theworking principle and with some advanced mechanism andengineering NNmethods have established a strong footprintin many problem-solving issues Figure 12 shows the basicworking principle of NN techniques
In the NN model the input data X = 1199090 1199091 119909119873 isfirst multiplied by the weight dataW = 1199080 1199081 119908119873 andthen the output is calculated using
Y = g (sum) wheresum = W sdot X (2)
Function g is known as the activation function Thisfunction can be any threshold value or Sigmoid or hyperbolicand so forth In the early stages feed-forwardNeuralNetworktechniques were introduced [57] lately the backpropagationmethod has been invented to utilize the error information toimprove the system performance [58 59]
The history of breast image classification by NN is a longone To the best of my knowledge a lot of the pioneer work
yg
x0
x1
xNminus1
xN
w0
w1
wNminus1
wN
Figure 12Working principle of a simpleNeuralNetwork technique
was performed by Dawson et al in 1991 [60] Since then NNhas been utilized as one of the strong tools for breast imageclassification We have summarized some of the work relatedto NN and breast image classification in Tables 5 6 and 7
312 Deep Neural Network Deep Neural Network (DNN) isa state-of-the-art concept where conventional NN techniqueshave been utilized with advanced engineering It is foundthat conventional NNs have difficulties in solving complexproblems whereas DNNs solve them with utmost PrecisionHowever DNNs suffer from more time and computationalcomplexity than the conventional NN
Convolutional Neural Network A CNN model is the combi-nation of a few intermediate mathematical structures Thisintermediatemathematical structure creates or helps to createdifferent layers
(i) Convolutional Layer Among all the other layers theconvolutional layer is considered as the most important partfor a CNN model and can be considered as the backbone of
Computational and Mathematical Methods in Medicine 9
Table 5 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Rajakeerthana et al [42] (1) GLCM GLDM SRDMNGLCM GLRM Mammogram 322 (1)The classifier achieved 9920
Accuracy
Lessa and Marengoni [43](1)Mean Median StandardDeviation Skewness KurtosisEntropy Range
Wan et al [44] (1) ALBP (2) BBLBP OCM 46(1) Achieved Sensitivity and Specificityare 100 and 8520 respectively(2) ROC value obtained 0959
Chen et al [40] (1) 19 BI-RADS features havebeen used Ultrasound 238
(1) Chi squared method has beenutilized for the feature selection(2) Achieved Accuracy Sensitivity andSpecificity are 9610 9670 and9570 respectively
de Lima et al [45] (1) Total 416 features have beenused Mammogram 355
(1)Multiresolution wavelet and Zernikemoment have been utilized for thefeature extraction
Abirami et al [46](1) 12 statistical measures such asMean Median and Max havebeen utilized as the features
Mammogram 322
(1)Wavelet transform has been utilizedfor the feature extraction(2)The achieved Accuracy Sensitivityand Specificity are 9550 9500 and9600 respectively
El Atlas et al [47] (1) 13 morphological featureshave been utilized Mammogram 410
(1) Firstly the edge information hasbeen utilized for the mass segmentationand then the morphological featureswere extracted(2) Achieved best Accuracy is 975
Table 6 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Alharbi et al [48] (1) 49 features havebeen utilized Mammogram 1100
(1) Five feature selection methods Fisher scoreMinimum Redundancy-Maximum Relevance Relief-fSequential Forward Feature Selection and GeneticAlgorithm have been used(2) Achieved Accuracy Sensitivity and specificity are9420 9836 and 9927 respectively
Peng et al [49](1)Haralick andTamura features havebeen utilized
Mammogram 322
(1) Feature reduction has been performed byRough-Set theory and selected 5 prioritized features(2)The best Accuracy Sensitivity and Specificityachieved were 9600 9860 and 8930
Jalalian et al [50] (1) GLCM Mammogram(1)The obtained classifier Accuracy Sensitivity andSpecificity are 9520 9240 and 9800respectively(2) Compactness
Li et al [51](1) Four featurevectors have beencalculated
Mammogram 322
(1) 2D contour of breast mass in mammography hasbeen converted into 1D signature(2) NN techniques achieved Accuracy is 9960 whenRMS slope is utilized
Chen et al [52] (1) Autocorrelationfeatures Ultrasound 242 (1)The overall achieved Accuracy Sensitivity and
Specificity are 9500 9800 and 93 respectively
Chen et al [53] (1) Autocorrelationfeatures Ultrasound 1020 (1)The obtained ROC area is 09840 plusmn 00072
10 Computational and Mathematical Methods in Medicine
Table 7 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Chen et al [61]
(1) Variance Contrast of WaveletCoefficient Ultrasound 242 (1)The achieved ROC curve 09396 plusmn 00183(2) Autocorrelation of WaveletCoefficient
Silva et al [62](1) 22 different morphologicalfeatures such as convexity andlobulation have been utilized
Ultrasound mdash (1)The best obtained Accuracy and ROCcurve are 9698 and 098 respectively
Saritas [63](1) Age of patient (2)massshape (3)mass border (4)Massdensity (5) BIRADS Mammogram mdash
(1) Disease prediction rate is 905(2) Neural Network utilized 5 neurons ininput layers and one hidden layer
Lopez-Melendez etal [64]
(1) Area perimeter etc havebeen utilized Mammogram 322 (1)The achieved Sensitivity and Specificity
are 9629 and 9900 respectively
themodel A kernel of size119898times119899 is scanned through the inputdata for the convolutional operation which ensures the localconnectivity and weight sharing property
(ii) Stride and Padding In the convolutional operation afilter scans through the input matrices In each step howmuch position a kernel filter moves through the matrixis known as the stride By default stride keeps to 1 Withinappropriate selection of the stride the model can lose theborder information To overcome this issue themodel utilizesextra rows and columns at the end of the matrices and theseadded rows and columns contain all 0s This adding of extrarows and columns which contain only zero value is known aszero padding
(iii) Nonlinear Operation The output of each of the kerneloperations is passed through a rectifier function such as Rec-tified Linear Unit (ReLU) Leaky-ReLU TanH and SigmoidThe Sigmoid function can be defined as
120590 (119909) = 1(1 + expminus119909) (3)
and the tanh function can be defined as
tanh (119909) = (exp119909 minus expminus119909)(exp119909 + expminus119909) (4)
However the most effective rectifier is ReLU The ReLUmethod converts all the information into zero if it is less thanor equal to zero and passes all the other data as is shown inFigure 13
120590 (119909) = max (0 119909) (5)
Another important nonlinear function is Leaky-RelU
where 120572 is predetermined parameter which can be varied togive a better model
minus3 minus2 minus1 0 1 2 3
1
2
3
InputO
utpu
t
Figure 13 ReLU Operation
(iv) Subsampling Subsampling is the procedure of reducingthe dimensionality of each of the feature maps of a particularlayer this operation is also known as a pooling operationActually it reduces the amount of feature information fromthe overall data By doing so it reduces the overall computa-tional complexity of themodel To do this 119904times119904 patch units areutilized The two most popular pooling methods are
(a) Max-Pooling
(b) Average Pooling
In Max-Pooling only the maximum values within a partic-ular kernel size are selected for further calculation Consideran example of a 16 times 16 image as shown in Figure 14 A 2 by2 kernel is applied to the whole image 4 blocks in total andproduces a 4 times 4 output image For each block of four valueswe have selected the maximum For instance from blocksone two three and four maximum values 4 40 13 and 8are selected respectively as they are the maximum in thatblock For the Average Pooling operation each kernel givesthe output as average
(v) Dropout Regularization of the weight can reduce theoutfitting problem Randomly removing some neurons can
Computational and Mathematical Methods in Medicine 11
Figure 15 Work-flow of a Convolutional Neural Network
regularize the overfilling problem The technique of ran-domly removing neurons from the network is known asdropout
(vi) Soft-Max Layer This layer contains normalized expo-nential functions to calculate the loss function for the dataclassification
Figure 15 shows a generalized CNN model for the imageclassificationAll the neurons of themost immediate layer of afully connected layer are completely connected with the fullyconnected layer like a conventional Neural Network Let119891119897minus1119895represent the 119895th feature map at the layer 119897minus1The 119895th featuremap at the layer 119897 can be represented as
where119873119897minus119897 represents the number of featuremaps at the 119897minus1thlayer 119896119894119895 represents the kernel function and 119887119897119895 represents thebias at 119897 where 120590 performs a nonlinear function operationThe layer before the Soft-Max Layer can be represented as
Let 119901 = 1 represent Benign class and 119901 = 2 represent theMalignant class The cross-entropy loss of the above functioncan be calculated as
119871119901 = minus ln (119910119901) (10)
Whichever group experiences a large loss value themodel will consider the other group as predicted class
A difficult part of working on DNN is that it requiresa specialized software package for the data analysis Fewresearch groups have been working on how effectively datacan be analyzed by DNN from different perspectives and thedemand Table 8 summarizes some of the software which isavailable for DNN analysis
The history of the CNN and its use for biomedical imageanalysis is a long one Fukushima first introduced a CNNnamed ldquonecognitronrdquo which has the ability to recognizestimulus patterns with a few shifting variances [113] Tothe best of our knowledge Wu et al first classified a setof mammogram images into malignant and benign classesusing a CNN model [78] In their proposed model they onlyutilized one hidden layer After that in 1996 Sahiner et alutilized CNNmodel to classify mass and normal breast tissueand achieved ROC scores of 087 [79] In 2002 Lo et alutilized aMultiple Circular Path CNN (MCPCNN) for tumoridentification from mammogram images and obtained ROCscores of around 089 After an absence of investigation ofthe CNN model this model regained its momentum afterthe work of Krizhevsky et al [114] Their proposed model isknown as AlexNet After this work a revolutionary change
12 Computational and Mathematical Methods in Medicine
Table 8 Available software for deep learning analysis
Software Interface and backend Provider
Caffe [65 66] Python MATLAB C++ Berkeley Vision and Learning CentreUniversity of California Berkeley
Torch [67] C LuaJIT
MatConvNet [68 69] MATLAB C Visual Geometry Group Department ofEngineering University of Oxford
Theano [70 71] Python Montreal Institute for Learning AlgorithmsUniversity of Montreal
TensorFlows [72] C++ Python GoogleCNTK [73] C++ MicrosoftKeras [74] Theano Tensor Flow MITdl4j [75] Java Skymind Engineering
DeeBNET [76 77] MATLAB Information Technology DepartmentAmirkabir University of Technology
has been achieved in the image classification and analysisfield As an advanced engineering of the AlexNet the papertitled ldquoGoing Deeper with Convolutionsrdquo by Szegedy [115]introduced the GoogleNet model This model contains amuch deeper network than AlexNet Sequentially ResNet[116] Inception [117] Inception-v4 Inception-ResNet [118]and a few other models have recently been introduced
Later directly or with some advanced modificationthese DNN models have been adapted for biomedical imageanalysis In 2015 Fonseca et al [81] classified breast densityusing CNN techniques CNN requires a sufficient amountof data to train the system It is always very difficult tofind a sufficient amount of medical data for training a CNNmodel A pretrained CNN model with some fine tuning canbe used rather than create a model from scratch [119] Theauthors of [119] did not perform their experiments on a breastcancer image dataset however they have performed theirexperiments on three different medical datasets with layer-wise training and claimed that ldquoretrained CNN along withadequate training can provide better or at least the sameamount of performancerdquo
The Deep Belief Network (DBN) is another branch of theDeep Neural Network which mainly consists of RestrictedBoltzmann Machine (RBM) techniques The DBN methodwas first utilized for supervised image classification by Liu etal [120] After that Abdel-Zaher and Eldeib utilized the DBNmethod for breast image classification [121] This field is stillnot fully explored for breast image classification yet Zhanget al utilized both RBM and Point-Wise Gated RBM (PRBM)for shear-wave electrography image classification where thedataset contains 227 images [97]Their achieved classificationAccuracy Sensitivity and Specificity are 9340 8860 and9710 respectively Tables 9 10 and 11 have summarized themost recent work for breast image classification along withsome pioneer work on CNN
313 Logic Based Algorithm A Logic Based algorithm isa very popular and effective classification method whichfollows the tree structure principle and logical argument asshown in Figure 16 This algorithm classifies instances based
on the featurersquos values Along with other criteria a decision-tree based algorithm contains the following features
(i) Root node a root node contains no incoming nodeand it may or may not contain any outgoing edge
(ii) Splitting splitting is the process of subdividing a set ofcases into a particular group Normally the followingcriteria are maintained for the splitting
(a) information gain(b) Gini index(c) chi squared
(iii) Decision node(iv) Leafterminal node this kind of node has exactly one
incoming edge and no outgoing edgeThe tree alwaysterminates here with a decision
(v) Pruning pruning is a process of removing subtreesfrom the tree Pruning performs to reduce the over-fitting problem Two kinds of pruning techniques areavailable
(a) prepruning(b) postpruning
Among all the tree based algorithms IterativeDichotomiser 3 (ID3) can be considered as a pioneerproposed by Quinlan [149] The problem of the ID3algorithm is to find the optimal solution which is very muchprone towards overfitting To overcome the limitation of theID3 algorithm the C45 algorithm has been introduced byQuinlan [150] where a pruning method has been introducedto control the overfitting problem Pritom et al [151] classifiedthe Wisconsin breast dataset where they utilized 35 featuresThey have obtained 7630 Accuracy 7510 False PositiveRate and ROC score 0745 when they ranked the featuresWithout ranking the features they obtained 7370Accuracy5070 False Positive Rate and ROC score value 5280 Asriet al [152] utilized the C45 algorithm for the Wisconsin
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Computational and Mathematical Methods in Medicine 3
Figure 3 Anatomy of the female breast images (for the National Cancer Institute 2011 Terese Winslow US Government has certain rights)
(a) (b)
(c) (d)
Figure 4 (a b) showmammogram benign and malignant images (examples of noninvasive image) and (c d) show histopathological benignand malignant images (examples of invasive image)
4 Computational and Mathematical Methods in Medicine
Table 1 Available breast image database for biomedical investigation
Figure 5 A very basic breast image classification model
outcome researchers always base their investigation on somewell-established image database Various organizations haveintroduced sets of images databases which are available toresearchers for further investigation Table 1 gives a few of theavailable image databases with some specifications
The image formats of the different databases are differentFew of the images contained images in JPEG format and fewdatabases contained DICOM-format data Here the MIASDDSM and Inbreast databases containmammogram imagesAccording to the Springer (httpwwwspringercom)Elsevier (httpswwwelseviercom) and IEEE (httpwwwieeexploreieeeorg) web sites researchers have mostlyutilized the MIAS and DDSM databases for the breast imageclassification research The number of conference paperspublished for the DDSM and MIAS databases is 110 and 168respectively with 82 journal papers published on DDSMdatabases and 136 journal papers published using the MIASdatabase We have verified these statistics on both Scopus(httpswwwscopuscom) and the Web of Science database(httpwwwwebofknowledgecom) Figure 6 shows thenumber of published breast image classification papers basedon the MIAS and DDSM database from the years 2000 to2017
Histopathological images provide valuable informationand are being intensively investigated by doctors for find-ing the current situation of the patient The TCGA-BRCAand BreakHis databases contain histopathological imagesResearch has been performed in a few experiments on thisdatabase too Among these two databases BreakHis is themost recent histopathological image database containing a
4 4 3 2 47
16
68 8
2319
37
19
38
4541
17
1 0 04 4 4
7 8 96
1215 14
2123
2826
12
05
101520253035404550
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
Freq
uenc
y
Year
MIASDDSM
Figure 6 Number of papers published based on MIAS and DDSMdatabases
total of 7909 images which have been collected from 82patients [6] So far around twenty research papers have beenpublished based on this database
22 Feature Extraction and Selection An important stepof the image classification is extracting the features fromthe images In the conventional image classification taskfeatures are crafted locally using some specific rules andcriteria However the-state-of-the-art Convolutional NeuralNetwork (CNN) techniques generally extract the featuresglobally using kernels and these Global Features have beenused for image classification Among the local featurestexture detector and statistical are being accepted as impor-tant features for breast image classification Texture featuresactually represent the low-level feature information of animage which providesmore detailed information of an imagethat might be possible from histogram information aloneMore specifically texture features provide the structural anddimensional information of the color as well as the intensity
Computational and Mathematical Methods in Medicine 5
Table 2 Feature descriptor
Feature category Feature description
Texture
Haralick texture features [7]
(1) Angular Second Moment (ASM) (2) Contrast (3) correlation (4) Sum of Squares of Variances (SSoV) (5) Inverseof Difference (IoD) (6) Sum of Average (SoA) (7) Sum of Variances (SoV) (8) Sum of Entropy (SoE) (9) Entropy(10) Difference of Variance (DoV) (11) Difference of Entropy (DoE) (12) Gray-Level Concurrence Matrix (GLCM)Tamura features [8](1) Coarseness (2) Contrast (3) directionality (4) line-likeness (5) roughness (6) regularityGlobal texture descriptor(1) Fractal dimension (FD) (2) Coarseness (3) Entropy (4) Spatial Gray-Level Statistics (SGLS) (5) Circular MoranAutocorrelation Function (CMAF)
Detector
Single scale detector(1)Moravecrsquos Detector (MD) [9] (2)Harris Detector (HD) [10] (3) Smallest Univalue Segment Assimilating Nucleus(SUSAN) [11] (4) Features from Accelerated Segment Test (FAST) [12 13] (5)Hessian Blob Detector (HBD) [14 15]Multiscale detector [8](1) Laplacian of Gaussian (LoG) [9 16] (2) Difference of Gaussian (DoG) Contrast [17] (3)Harris Laplace (HL) (4)Hessian Laplace (HeL) (5) Gabor-Wavelet Detector (GWD) [18]
Figure 7 Classification of features for breast image classification
of the image Breast Imaging-Reporting and Data System(BI-RADS) is a mammography image assessment techniquecontaining 6 categories normally assigned by the radiologistFeature detector actually provides information whether theparticular feature is available in the image or not Structuralfeatures provide information about the features structure andorientation such as the area Convex Hull and centroid Thiskind of information gives more detailed information aboutthe features In a cancer image it can provide the area ofthe nucleus or the centroid of the mass Mean Medianand Standard Deviation always provide some importantinformation on the dataset and their distribution This kindof features has been categorized as statistical features Thetotal hierarchy of the image feature extraction is resented inFigure 7 Tables 2 and 3 further summarize the local featuresin detail
Features which are extracted for classification do notalways carry the same importance Some features may evencontribute to degrading the classifier performance Priori-tization of the feature set can reduce the classifier modelcomplexity and so it can reduce the computational timeFeature set selection and prioritization can be classified intothree broad categories
(i) Filter the filter method selects features without eval-uating any classifier algorithm
(ii) Wrapper the wrapper method selects the feature setbased on the evaluation performance of a particularclassifier
(iii) Embedded the embeddedmethod takes advantage ofthe filter andwrappermethods for classifier construc-tion
6 Computational and Mathematical Methods in Medicine
Table 3 Feature descriptor
Feature category Feature descriptionStatistical (1)Mean (2)Median (3) Standard Deviation (4) Skewness (5) Kurtosis (6) Range
Descriptor
(1) Scale Invariant Feature Transform (SIFT) [17 19] (2) Gradient Location-Orientation Histogram (GLOH) [20] (3)Speeded-Up Robust Features Descriptor (SURF) [21ndash23] (4) Local Binary Pattern (LBP) [24ndash27] (5) Binary RobustIndependent Elementary Features (BRIEF) [28] (6)Weber Local Descriptor (WLD) [29 30] (7) Back Ground LocalBinary Pattern (BGLBP) [31] (8) Center-Symmetric Local Binary Pattern (CS-LBP) [32] (9) Second-OrderCenter-Symmetric Local Derivative Pattern (CS-LBP) [33] (10) Center-Symmetric Scale Invariant Local TernaryPatterns (CS-SILTP) [34] (11) Extended LBP or Circular LBP (E-LBP) [35] (12)Opponent Color Local Binary Pattern(OC-LBP) [36] (13) Original LBP(O-LBP) [25] (14) Spatial Extended Center-Symmetric Local Binary Pattern(SCS-LBP) [37] (15) Scale Invariant Local Ternary Pattern (SI-LTP) [38] (16) Variance-Based LBP (VAR-LBP) [24](17) eXtended Center-Symmetric Local Binary Pattern (XCS-LBP) (18) Average Local Binary Pattern (ALBP) (19)Block Based Local Binary Pattern (BBLBP) [39]
Figure 8 shows a generalized feature selection methodwhere we have further classified the filter method intoFisher Score Mutual Information Relief and chi squaremethods The embedded method has been classified intoBridge Regularization Lasso and Adaptive Lasso methodswhile the wrapper method has been classified to recursivefeature selection and sequential feature selection method
23 Classifier Model Based on the learning point of viewbreast image classification techniques can be categorized intothe following three classes [41]
These three classes can be split into Deep Neural Network(DNN) and conventional classifier (without DNN) and tosome further classes as in Table 4
24 Performance Measuring Parameter A Confusion Matrixis a two-dimensional table which is used to a give a visual
True
clas
s
Hypothesized class
True positive (A) False negative (B)
False positive (C) True negative (D)
Figure 9 Confusion Matrix
perception of classification experiments [54] The (119894 119895)thposition of the confusion table indicates the number of timesthat the 119894th object is classified as the 119895th object The diagonalof this matrix indicates the number of times the objects arecorrectly classified Figure 9 shows a graphical representationof a Confusion Matrix for the binary classification case
Computational and Mathematical Methods in Medicine 7
Table 4 A simplified hierarchy of classification
Learning technique Algorithm
Supervised
Conventional
(a) Logic based
(1) ID3 (2) C45 (3) bagging(4) random trees (5) Random Forest(6) boosting (7) advanced boosting(8) Extreme Boosting (XGBoosting)
(a) Self-training(b) Graph Based(c) S3V3(d) Multiview(e) Generative model
Among the different classification performance proper-ties this matrix will provide following parameters
(i) Recall is defined as Recall = TP(TP + FN)(ii) Precision is defined as Precision = TP(TP + FP)(iii) Specificity is defined as Specificity = TN(TN + FP)(iv) Accuracy is defined as ACC = (TP+TN)(TP+TN+
FP + FN)(v) F-1 score is defined as 1198651 = (2 times Recall)(2 times Recall +
FP + FN)(vi) Matthew Correlation Coefficient (MCC) MCC is a
performance parameter of a binary classifier in therange minus1 to +1 If the MCC values trend moretowards +1 the classifier gives a more accurate classi-fier and the opposite condition will occur if the valueof theMCC trend towards theminus1MCCcanbe definedas
MCC
= TP times TN minus FP times FNradic(TP + FP) (TP + FN) (TN + FP) (TN + FP) (1)
3 Performance of Different Classifier Modelon Breast Images Dataset
Based on Supervised Semisupervised and Unsupervisedmethods different research groups have been performedclassification operation on different image database In thissection we have summarized few of the works of breast imageclassification
31 Performance Based on Supervised Learning In super-vised learning a general hypothesis is established based onexternally supplied instances to produce future predictionFor the supervised classification task features are extractedor automatically crafted from the available dataset and eachsample is mapped to a dedicated class With the help of thefeatures and their levels a hypothesis is created Based on thehypothesis unknown data are classified [55]
Figure 10 represents an overall supervised classifier archi-tecture In general the whole dataset is split into trainingand testing parts To validate the data some time dataare also split into a validation part as well After the datasplitting themost important part is to find out the appropriatefeatures to classify the data with the utmost AccuracyFinding the features can be classified into two categorieslocally and globally crafted Locally crafted means that thismethod requires a hand-held exercise to find out the featureswhereas globally craftedmeans that a kernelmethod has beenintroduced for the feature extraction Handcrafted featurescan be prioritized whereas Global Feature selection does nothave this luxury
311 Conventional Neural Network The Neural Network(NN) concept comes from the working principle of thehuman brain A biological neuron consists of the followingfour parts
8 Computational and Mathematical Methods in Medicine
Classifier model
Imagedatabase
Traintestdata splitting Locally
craftedGloballycrafted
Hand crafting
Kernel basedcrafting
Featureprioritization
Conventionalclassifier
DNNclassifier
Evaluationmatrix
Classifieddata
Feature collection
Ensemble learning
Figure 10 A generalized supervised classifier model
Nucleus
Axon
Cell body
Dendrites
Figure 11 A model of a biological neuron
Dendrites collect signals and axons carry the signal to thenext dendrite after processing by the cell body as shown inFigure 11 Using the neuronworking principle the perceptronmodel was proposed by Rosenblatt in 1957 [56] A single-layer perceptron linearly combines the input signal and givesa decision based on a threshold function Based on theworking principle and with some advanced mechanism andengineering NNmethods have established a strong footprintin many problem-solving issues Figure 12 shows the basicworking principle of NN techniques
In the NN model the input data X = 1199090 1199091 119909119873 isfirst multiplied by the weight dataW = 1199080 1199081 119908119873 andthen the output is calculated using
Y = g (sum) wheresum = W sdot X (2)
Function g is known as the activation function Thisfunction can be any threshold value or Sigmoid or hyperbolicand so forth In the early stages feed-forwardNeuralNetworktechniques were introduced [57] lately the backpropagationmethod has been invented to utilize the error information toimprove the system performance [58 59]
The history of breast image classification by NN is a longone To the best of my knowledge a lot of the pioneer work
yg
x0
x1
xNminus1
xN
w0
w1
wNminus1
wN
Figure 12Working principle of a simpleNeuralNetwork technique
was performed by Dawson et al in 1991 [60] Since then NNhas been utilized as one of the strong tools for breast imageclassification We have summarized some of the work relatedto NN and breast image classification in Tables 5 6 and 7
312 Deep Neural Network Deep Neural Network (DNN) isa state-of-the-art concept where conventional NN techniqueshave been utilized with advanced engineering It is foundthat conventional NNs have difficulties in solving complexproblems whereas DNNs solve them with utmost PrecisionHowever DNNs suffer from more time and computationalcomplexity than the conventional NN
Convolutional Neural Network A CNN model is the combi-nation of a few intermediate mathematical structures Thisintermediatemathematical structure creates or helps to createdifferent layers
(i) Convolutional Layer Among all the other layers theconvolutional layer is considered as the most important partfor a CNN model and can be considered as the backbone of
Computational and Mathematical Methods in Medicine 9
Table 5 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Rajakeerthana et al [42] (1) GLCM GLDM SRDMNGLCM GLRM Mammogram 322 (1)The classifier achieved 9920
Accuracy
Lessa and Marengoni [43](1)Mean Median StandardDeviation Skewness KurtosisEntropy Range
Wan et al [44] (1) ALBP (2) BBLBP OCM 46(1) Achieved Sensitivity and Specificityare 100 and 8520 respectively(2) ROC value obtained 0959
Chen et al [40] (1) 19 BI-RADS features havebeen used Ultrasound 238
(1) Chi squared method has beenutilized for the feature selection(2) Achieved Accuracy Sensitivity andSpecificity are 9610 9670 and9570 respectively
de Lima et al [45] (1) Total 416 features have beenused Mammogram 355
(1)Multiresolution wavelet and Zernikemoment have been utilized for thefeature extraction
Abirami et al [46](1) 12 statistical measures such asMean Median and Max havebeen utilized as the features
Mammogram 322
(1)Wavelet transform has been utilizedfor the feature extraction(2)The achieved Accuracy Sensitivityand Specificity are 9550 9500 and9600 respectively
El Atlas et al [47] (1) 13 morphological featureshave been utilized Mammogram 410
(1) Firstly the edge information hasbeen utilized for the mass segmentationand then the morphological featureswere extracted(2) Achieved best Accuracy is 975
Table 6 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Alharbi et al [48] (1) 49 features havebeen utilized Mammogram 1100
(1) Five feature selection methods Fisher scoreMinimum Redundancy-Maximum Relevance Relief-fSequential Forward Feature Selection and GeneticAlgorithm have been used(2) Achieved Accuracy Sensitivity and specificity are9420 9836 and 9927 respectively
Peng et al [49](1)Haralick andTamura features havebeen utilized
Mammogram 322
(1) Feature reduction has been performed byRough-Set theory and selected 5 prioritized features(2)The best Accuracy Sensitivity and Specificityachieved were 9600 9860 and 8930
Jalalian et al [50] (1) GLCM Mammogram(1)The obtained classifier Accuracy Sensitivity andSpecificity are 9520 9240 and 9800respectively(2) Compactness
Li et al [51](1) Four featurevectors have beencalculated
Mammogram 322
(1) 2D contour of breast mass in mammography hasbeen converted into 1D signature(2) NN techniques achieved Accuracy is 9960 whenRMS slope is utilized
Chen et al [52] (1) Autocorrelationfeatures Ultrasound 242 (1)The overall achieved Accuracy Sensitivity and
Specificity are 9500 9800 and 93 respectively
Chen et al [53] (1) Autocorrelationfeatures Ultrasound 1020 (1)The obtained ROC area is 09840 plusmn 00072
10 Computational and Mathematical Methods in Medicine
Table 7 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Chen et al [61]
(1) Variance Contrast of WaveletCoefficient Ultrasound 242 (1)The achieved ROC curve 09396 plusmn 00183(2) Autocorrelation of WaveletCoefficient
Silva et al [62](1) 22 different morphologicalfeatures such as convexity andlobulation have been utilized
Ultrasound mdash (1)The best obtained Accuracy and ROCcurve are 9698 and 098 respectively
Saritas [63](1) Age of patient (2)massshape (3)mass border (4)Massdensity (5) BIRADS Mammogram mdash
(1) Disease prediction rate is 905(2) Neural Network utilized 5 neurons ininput layers and one hidden layer
Lopez-Melendez etal [64]
(1) Area perimeter etc havebeen utilized Mammogram 322 (1)The achieved Sensitivity and Specificity
are 9629 and 9900 respectively
themodel A kernel of size119898times119899 is scanned through the inputdata for the convolutional operation which ensures the localconnectivity and weight sharing property
(ii) Stride and Padding In the convolutional operation afilter scans through the input matrices In each step howmuch position a kernel filter moves through the matrixis known as the stride By default stride keeps to 1 Withinappropriate selection of the stride the model can lose theborder information To overcome this issue themodel utilizesextra rows and columns at the end of the matrices and theseadded rows and columns contain all 0s This adding of extrarows and columns which contain only zero value is known aszero padding
(iii) Nonlinear Operation The output of each of the kerneloperations is passed through a rectifier function such as Rec-tified Linear Unit (ReLU) Leaky-ReLU TanH and SigmoidThe Sigmoid function can be defined as
120590 (119909) = 1(1 + expminus119909) (3)
and the tanh function can be defined as
tanh (119909) = (exp119909 minus expminus119909)(exp119909 + expminus119909) (4)
However the most effective rectifier is ReLU The ReLUmethod converts all the information into zero if it is less thanor equal to zero and passes all the other data as is shown inFigure 13
120590 (119909) = max (0 119909) (5)
Another important nonlinear function is Leaky-RelU
where 120572 is predetermined parameter which can be varied togive a better model
minus3 minus2 minus1 0 1 2 3
1
2
3
InputO
utpu
t
Figure 13 ReLU Operation
(iv) Subsampling Subsampling is the procedure of reducingthe dimensionality of each of the feature maps of a particularlayer this operation is also known as a pooling operationActually it reduces the amount of feature information fromthe overall data By doing so it reduces the overall computa-tional complexity of themodel To do this 119904times119904 patch units areutilized The two most popular pooling methods are
(a) Max-Pooling
(b) Average Pooling
In Max-Pooling only the maximum values within a partic-ular kernel size are selected for further calculation Consideran example of a 16 times 16 image as shown in Figure 14 A 2 by2 kernel is applied to the whole image 4 blocks in total andproduces a 4 times 4 output image For each block of four valueswe have selected the maximum For instance from blocksone two three and four maximum values 4 40 13 and 8are selected respectively as they are the maximum in thatblock For the Average Pooling operation each kernel givesthe output as average
(v) Dropout Regularization of the weight can reduce theoutfitting problem Randomly removing some neurons can
Computational and Mathematical Methods in Medicine 11
Figure 15 Work-flow of a Convolutional Neural Network
regularize the overfilling problem The technique of ran-domly removing neurons from the network is known asdropout
(vi) Soft-Max Layer This layer contains normalized expo-nential functions to calculate the loss function for the dataclassification
Figure 15 shows a generalized CNN model for the imageclassificationAll the neurons of themost immediate layer of afully connected layer are completely connected with the fullyconnected layer like a conventional Neural Network Let119891119897minus1119895represent the 119895th feature map at the layer 119897minus1The 119895th featuremap at the layer 119897 can be represented as
where119873119897minus119897 represents the number of featuremaps at the 119897minus1thlayer 119896119894119895 represents the kernel function and 119887119897119895 represents thebias at 119897 where 120590 performs a nonlinear function operationThe layer before the Soft-Max Layer can be represented as
Let 119901 = 1 represent Benign class and 119901 = 2 represent theMalignant class The cross-entropy loss of the above functioncan be calculated as
119871119901 = minus ln (119910119901) (10)
Whichever group experiences a large loss value themodel will consider the other group as predicted class
A difficult part of working on DNN is that it requiresa specialized software package for the data analysis Fewresearch groups have been working on how effectively datacan be analyzed by DNN from different perspectives and thedemand Table 8 summarizes some of the software which isavailable for DNN analysis
The history of the CNN and its use for biomedical imageanalysis is a long one Fukushima first introduced a CNNnamed ldquonecognitronrdquo which has the ability to recognizestimulus patterns with a few shifting variances [113] Tothe best of our knowledge Wu et al first classified a setof mammogram images into malignant and benign classesusing a CNN model [78] In their proposed model they onlyutilized one hidden layer After that in 1996 Sahiner et alutilized CNNmodel to classify mass and normal breast tissueand achieved ROC scores of 087 [79] In 2002 Lo et alutilized aMultiple Circular Path CNN (MCPCNN) for tumoridentification from mammogram images and obtained ROCscores of around 089 After an absence of investigation ofthe CNN model this model regained its momentum afterthe work of Krizhevsky et al [114] Their proposed model isknown as AlexNet After this work a revolutionary change
12 Computational and Mathematical Methods in Medicine
Table 8 Available software for deep learning analysis
Software Interface and backend Provider
Caffe [65 66] Python MATLAB C++ Berkeley Vision and Learning CentreUniversity of California Berkeley
Torch [67] C LuaJIT
MatConvNet [68 69] MATLAB C Visual Geometry Group Department ofEngineering University of Oxford
Theano [70 71] Python Montreal Institute for Learning AlgorithmsUniversity of Montreal
TensorFlows [72] C++ Python GoogleCNTK [73] C++ MicrosoftKeras [74] Theano Tensor Flow MITdl4j [75] Java Skymind Engineering
DeeBNET [76 77] MATLAB Information Technology DepartmentAmirkabir University of Technology
has been achieved in the image classification and analysisfield As an advanced engineering of the AlexNet the papertitled ldquoGoing Deeper with Convolutionsrdquo by Szegedy [115]introduced the GoogleNet model This model contains amuch deeper network than AlexNet Sequentially ResNet[116] Inception [117] Inception-v4 Inception-ResNet [118]and a few other models have recently been introduced
Later directly or with some advanced modificationthese DNN models have been adapted for biomedical imageanalysis In 2015 Fonseca et al [81] classified breast densityusing CNN techniques CNN requires a sufficient amountof data to train the system It is always very difficult tofind a sufficient amount of medical data for training a CNNmodel A pretrained CNN model with some fine tuning canbe used rather than create a model from scratch [119] Theauthors of [119] did not perform their experiments on a breastcancer image dataset however they have performed theirexperiments on three different medical datasets with layer-wise training and claimed that ldquoretrained CNN along withadequate training can provide better or at least the sameamount of performancerdquo
The Deep Belief Network (DBN) is another branch of theDeep Neural Network which mainly consists of RestrictedBoltzmann Machine (RBM) techniques The DBN methodwas first utilized for supervised image classification by Liu etal [120] After that Abdel-Zaher and Eldeib utilized the DBNmethod for breast image classification [121] This field is stillnot fully explored for breast image classification yet Zhanget al utilized both RBM and Point-Wise Gated RBM (PRBM)for shear-wave electrography image classification where thedataset contains 227 images [97]Their achieved classificationAccuracy Sensitivity and Specificity are 9340 8860 and9710 respectively Tables 9 10 and 11 have summarized themost recent work for breast image classification along withsome pioneer work on CNN
313 Logic Based Algorithm A Logic Based algorithm isa very popular and effective classification method whichfollows the tree structure principle and logical argument asshown in Figure 16 This algorithm classifies instances based
on the featurersquos values Along with other criteria a decision-tree based algorithm contains the following features
(i) Root node a root node contains no incoming nodeand it may or may not contain any outgoing edge
(ii) Splitting splitting is the process of subdividing a set ofcases into a particular group Normally the followingcriteria are maintained for the splitting
(a) information gain(b) Gini index(c) chi squared
(iii) Decision node(iv) Leafterminal node this kind of node has exactly one
incoming edge and no outgoing edgeThe tree alwaysterminates here with a decision
(v) Pruning pruning is a process of removing subtreesfrom the tree Pruning performs to reduce the over-fitting problem Two kinds of pruning techniques areavailable
(a) prepruning(b) postpruning
Among all the tree based algorithms IterativeDichotomiser 3 (ID3) can be considered as a pioneerproposed by Quinlan [149] The problem of the ID3algorithm is to find the optimal solution which is very muchprone towards overfitting To overcome the limitation of theID3 algorithm the C45 algorithm has been introduced byQuinlan [150] where a pruning method has been introducedto control the overfitting problem Pritom et al [151] classifiedthe Wisconsin breast dataset where they utilized 35 featuresThey have obtained 7630 Accuracy 7510 False PositiveRate and ROC score 0745 when they ranked the featuresWithout ranking the features they obtained 7370Accuracy5070 False Positive Rate and ROC score value 5280 Asriet al [152] utilized the C45 algorithm for the Wisconsin
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Figure 5 A very basic breast image classification model
outcome researchers always base their investigation on somewell-established image database Various organizations haveintroduced sets of images databases which are available toresearchers for further investigation Table 1 gives a few of theavailable image databases with some specifications
The image formats of the different databases are differentFew of the images contained images in JPEG format and fewdatabases contained DICOM-format data Here the MIASDDSM and Inbreast databases containmammogram imagesAccording to the Springer (httpwwwspringercom)Elsevier (httpswwwelseviercom) and IEEE (httpwwwieeexploreieeeorg) web sites researchers have mostlyutilized the MIAS and DDSM databases for the breast imageclassification research The number of conference paperspublished for the DDSM and MIAS databases is 110 and 168respectively with 82 journal papers published on DDSMdatabases and 136 journal papers published using the MIASdatabase We have verified these statistics on both Scopus(httpswwwscopuscom) and the Web of Science database(httpwwwwebofknowledgecom) Figure 6 shows thenumber of published breast image classification papers basedon the MIAS and DDSM database from the years 2000 to2017
Histopathological images provide valuable informationand are being intensively investigated by doctors for find-ing the current situation of the patient The TCGA-BRCAand BreakHis databases contain histopathological imagesResearch has been performed in a few experiments on thisdatabase too Among these two databases BreakHis is themost recent histopathological image database containing a
4 4 3 2 47
16
68 8
2319
37
19
38
4541
17
1 0 04 4 4
7 8 96
1215 14
2123
2826
12
05
101520253035404550
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
Freq
uenc
y
Year
MIASDDSM
Figure 6 Number of papers published based on MIAS and DDSMdatabases
total of 7909 images which have been collected from 82patients [6] So far around twenty research papers have beenpublished based on this database
22 Feature Extraction and Selection An important stepof the image classification is extracting the features fromthe images In the conventional image classification taskfeatures are crafted locally using some specific rules andcriteria However the-state-of-the-art Convolutional NeuralNetwork (CNN) techniques generally extract the featuresglobally using kernels and these Global Features have beenused for image classification Among the local featurestexture detector and statistical are being accepted as impor-tant features for breast image classification Texture featuresactually represent the low-level feature information of animage which providesmore detailed information of an imagethat might be possible from histogram information aloneMore specifically texture features provide the structural anddimensional information of the color as well as the intensity
Computational and Mathematical Methods in Medicine 5
Table 2 Feature descriptor
Feature category Feature description
Texture
Haralick texture features [7]
(1) Angular Second Moment (ASM) (2) Contrast (3) correlation (4) Sum of Squares of Variances (SSoV) (5) Inverseof Difference (IoD) (6) Sum of Average (SoA) (7) Sum of Variances (SoV) (8) Sum of Entropy (SoE) (9) Entropy(10) Difference of Variance (DoV) (11) Difference of Entropy (DoE) (12) Gray-Level Concurrence Matrix (GLCM)Tamura features [8](1) Coarseness (2) Contrast (3) directionality (4) line-likeness (5) roughness (6) regularityGlobal texture descriptor(1) Fractal dimension (FD) (2) Coarseness (3) Entropy (4) Spatial Gray-Level Statistics (SGLS) (5) Circular MoranAutocorrelation Function (CMAF)
Detector
Single scale detector(1)Moravecrsquos Detector (MD) [9] (2)Harris Detector (HD) [10] (3) Smallest Univalue Segment Assimilating Nucleus(SUSAN) [11] (4) Features from Accelerated Segment Test (FAST) [12 13] (5)Hessian Blob Detector (HBD) [14 15]Multiscale detector [8](1) Laplacian of Gaussian (LoG) [9 16] (2) Difference of Gaussian (DoG) Contrast [17] (3)Harris Laplace (HL) (4)Hessian Laplace (HeL) (5) Gabor-Wavelet Detector (GWD) [18]
Figure 7 Classification of features for breast image classification
of the image Breast Imaging-Reporting and Data System(BI-RADS) is a mammography image assessment techniquecontaining 6 categories normally assigned by the radiologistFeature detector actually provides information whether theparticular feature is available in the image or not Structuralfeatures provide information about the features structure andorientation such as the area Convex Hull and centroid Thiskind of information gives more detailed information aboutthe features In a cancer image it can provide the area ofthe nucleus or the centroid of the mass Mean Medianand Standard Deviation always provide some importantinformation on the dataset and their distribution This kindof features has been categorized as statistical features Thetotal hierarchy of the image feature extraction is resented inFigure 7 Tables 2 and 3 further summarize the local featuresin detail
Features which are extracted for classification do notalways carry the same importance Some features may evencontribute to degrading the classifier performance Priori-tization of the feature set can reduce the classifier modelcomplexity and so it can reduce the computational timeFeature set selection and prioritization can be classified intothree broad categories
(i) Filter the filter method selects features without eval-uating any classifier algorithm
(ii) Wrapper the wrapper method selects the feature setbased on the evaluation performance of a particularclassifier
(iii) Embedded the embeddedmethod takes advantage ofthe filter andwrappermethods for classifier construc-tion
6 Computational and Mathematical Methods in Medicine
Table 3 Feature descriptor
Feature category Feature descriptionStatistical (1)Mean (2)Median (3) Standard Deviation (4) Skewness (5) Kurtosis (6) Range
Descriptor
(1) Scale Invariant Feature Transform (SIFT) [17 19] (2) Gradient Location-Orientation Histogram (GLOH) [20] (3)Speeded-Up Robust Features Descriptor (SURF) [21ndash23] (4) Local Binary Pattern (LBP) [24ndash27] (5) Binary RobustIndependent Elementary Features (BRIEF) [28] (6)Weber Local Descriptor (WLD) [29 30] (7) Back Ground LocalBinary Pattern (BGLBP) [31] (8) Center-Symmetric Local Binary Pattern (CS-LBP) [32] (9) Second-OrderCenter-Symmetric Local Derivative Pattern (CS-LBP) [33] (10) Center-Symmetric Scale Invariant Local TernaryPatterns (CS-SILTP) [34] (11) Extended LBP or Circular LBP (E-LBP) [35] (12)Opponent Color Local Binary Pattern(OC-LBP) [36] (13) Original LBP(O-LBP) [25] (14) Spatial Extended Center-Symmetric Local Binary Pattern(SCS-LBP) [37] (15) Scale Invariant Local Ternary Pattern (SI-LTP) [38] (16) Variance-Based LBP (VAR-LBP) [24](17) eXtended Center-Symmetric Local Binary Pattern (XCS-LBP) (18) Average Local Binary Pattern (ALBP) (19)Block Based Local Binary Pattern (BBLBP) [39]
Figure 8 shows a generalized feature selection methodwhere we have further classified the filter method intoFisher Score Mutual Information Relief and chi squaremethods The embedded method has been classified intoBridge Regularization Lasso and Adaptive Lasso methodswhile the wrapper method has been classified to recursivefeature selection and sequential feature selection method
23 Classifier Model Based on the learning point of viewbreast image classification techniques can be categorized intothe following three classes [41]
These three classes can be split into Deep Neural Network(DNN) and conventional classifier (without DNN) and tosome further classes as in Table 4
24 Performance Measuring Parameter A Confusion Matrixis a two-dimensional table which is used to a give a visual
True
clas
s
Hypothesized class
True positive (A) False negative (B)
False positive (C) True negative (D)
Figure 9 Confusion Matrix
perception of classification experiments [54] The (119894 119895)thposition of the confusion table indicates the number of timesthat the 119894th object is classified as the 119895th object The diagonalof this matrix indicates the number of times the objects arecorrectly classified Figure 9 shows a graphical representationof a Confusion Matrix for the binary classification case
Computational and Mathematical Methods in Medicine 7
Table 4 A simplified hierarchy of classification
Learning technique Algorithm
Supervised
Conventional
(a) Logic based
(1) ID3 (2) C45 (3) bagging(4) random trees (5) Random Forest(6) boosting (7) advanced boosting(8) Extreme Boosting (XGBoosting)
(a) Self-training(b) Graph Based(c) S3V3(d) Multiview(e) Generative model
Among the different classification performance proper-ties this matrix will provide following parameters
(i) Recall is defined as Recall = TP(TP + FN)(ii) Precision is defined as Precision = TP(TP + FP)(iii) Specificity is defined as Specificity = TN(TN + FP)(iv) Accuracy is defined as ACC = (TP+TN)(TP+TN+
FP + FN)(v) F-1 score is defined as 1198651 = (2 times Recall)(2 times Recall +
FP + FN)(vi) Matthew Correlation Coefficient (MCC) MCC is a
performance parameter of a binary classifier in therange minus1 to +1 If the MCC values trend moretowards +1 the classifier gives a more accurate classi-fier and the opposite condition will occur if the valueof theMCC trend towards theminus1MCCcanbe definedas
MCC
= TP times TN minus FP times FNradic(TP + FP) (TP + FN) (TN + FP) (TN + FP) (1)
3 Performance of Different Classifier Modelon Breast Images Dataset
Based on Supervised Semisupervised and Unsupervisedmethods different research groups have been performedclassification operation on different image database In thissection we have summarized few of the works of breast imageclassification
31 Performance Based on Supervised Learning In super-vised learning a general hypothesis is established based onexternally supplied instances to produce future predictionFor the supervised classification task features are extractedor automatically crafted from the available dataset and eachsample is mapped to a dedicated class With the help of thefeatures and their levels a hypothesis is created Based on thehypothesis unknown data are classified [55]
Figure 10 represents an overall supervised classifier archi-tecture In general the whole dataset is split into trainingand testing parts To validate the data some time dataare also split into a validation part as well After the datasplitting themost important part is to find out the appropriatefeatures to classify the data with the utmost AccuracyFinding the features can be classified into two categorieslocally and globally crafted Locally crafted means that thismethod requires a hand-held exercise to find out the featureswhereas globally craftedmeans that a kernelmethod has beenintroduced for the feature extraction Handcrafted featurescan be prioritized whereas Global Feature selection does nothave this luxury
311 Conventional Neural Network The Neural Network(NN) concept comes from the working principle of thehuman brain A biological neuron consists of the followingfour parts
8 Computational and Mathematical Methods in Medicine
Classifier model
Imagedatabase
Traintestdata splitting Locally
craftedGloballycrafted
Hand crafting
Kernel basedcrafting
Featureprioritization
Conventionalclassifier
DNNclassifier
Evaluationmatrix
Classifieddata
Feature collection
Ensemble learning
Figure 10 A generalized supervised classifier model
Nucleus
Axon
Cell body
Dendrites
Figure 11 A model of a biological neuron
Dendrites collect signals and axons carry the signal to thenext dendrite after processing by the cell body as shown inFigure 11 Using the neuronworking principle the perceptronmodel was proposed by Rosenblatt in 1957 [56] A single-layer perceptron linearly combines the input signal and givesa decision based on a threshold function Based on theworking principle and with some advanced mechanism andengineering NNmethods have established a strong footprintin many problem-solving issues Figure 12 shows the basicworking principle of NN techniques
In the NN model the input data X = 1199090 1199091 119909119873 isfirst multiplied by the weight dataW = 1199080 1199081 119908119873 andthen the output is calculated using
Y = g (sum) wheresum = W sdot X (2)
Function g is known as the activation function Thisfunction can be any threshold value or Sigmoid or hyperbolicand so forth In the early stages feed-forwardNeuralNetworktechniques were introduced [57] lately the backpropagationmethod has been invented to utilize the error information toimprove the system performance [58 59]
The history of breast image classification by NN is a longone To the best of my knowledge a lot of the pioneer work
yg
x0
x1
xNminus1
xN
w0
w1
wNminus1
wN
Figure 12Working principle of a simpleNeuralNetwork technique
was performed by Dawson et al in 1991 [60] Since then NNhas been utilized as one of the strong tools for breast imageclassification We have summarized some of the work relatedto NN and breast image classification in Tables 5 6 and 7
312 Deep Neural Network Deep Neural Network (DNN) isa state-of-the-art concept where conventional NN techniqueshave been utilized with advanced engineering It is foundthat conventional NNs have difficulties in solving complexproblems whereas DNNs solve them with utmost PrecisionHowever DNNs suffer from more time and computationalcomplexity than the conventional NN
Convolutional Neural Network A CNN model is the combi-nation of a few intermediate mathematical structures Thisintermediatemathematical structure creates or helps to createdifferent layers
(i) Convolutional Layer Among all the other layers theconvolutional layer is considered as the most important partfor a CNN model and can be considered as the backbone of
Computational and Mathematical Methods in Medicine 9
Table 5 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Rajakeerthana et al [42] (1) GLCM GLDM SRDMNGLCM GLRM Mammogram 322 (1)The classifier achieved 9920
Accuracy
Lessa and Marengoni [43](1)Mean Median StandardDeviation Skewness KurtosisEntropy Range
Wan et al [44] (1) ALBP (2) BBLBP OCM 46(1) Achieved Sensitivity and Specificityare 100 and 8520 respectively(2) ROC value obtained 0959
Chen et al [40] (1) 19 BI-RADS features havebeen used Ultrasound 238
(1) Chi squared method has beenutilized for the feature selection(2) Achieved Accuracy Sensitivity andSpecificity are 9610 9670 and9570 respectively
de Lima et al [45] (1) Total 416 features have beenused Mammogram 355
(1)Multiresolution wavelet and Zernikemoment have been utilized for thefeature extraction
Abirami et al [46](1) 12 statistical measures such asMean Median and Max havebeen utilized as the features
Mammogram 322
(1)Wavelet transform has been utilizedfor the feature extraction(2)The achieved Accuracy Sensitivityand Specificity are 9550 9500 and9600 respectively
El Atlas et al [47] (1) 13 morphological featureshave been utilized Mammogram 410
(1) Firstly the edge information hasbeen utilized for the mass segmentationand then the morphological featureswere extracted(2) Achieved best Accuracy is 975
Table 6 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Alharbi et al [48] (1) 49 features havebeen utilized Mammogram 1100
(1) Five feature selection methods Fisher scoreMinimum Redundancy-Maximum Relevance Relief-fSequential Forward Feature Selection and GeneticAlgorithm have been used(2) Achieved Accuracy Sensitivity and specificity are9420 9836 and 9927 respectively
Peng et al [49](1)Haralick andTamura features havebeen utilized
Mammogram 322
(1) Feature reduction has been performed byRough-Set theory and selected 5 prioritized features(2)The best Accuracy Sensitivity and Specificityachieved were 9600 9860 and 8930
Jalalian et al [50] (1) GLCM Mammogram(1)The obtained classifier Accuracy Sensitivity andSpecificity are 9520 9240 and 9800respectively(2) Compactness
Li et al [51](1) Four featurevectors have beencalculated
Mammogram 322
(1) 2D contour of breast mass in mammography hasbeen converted into 1D signature(2) NN techniques achieved Accuracy is 9960 whenRMS slope is utilized
Chen et al [52] (1) Autocorrelationfeatures Ultrasound 242 (1)The overall achieved Accuracy Sensitivity and
Specificity are 9500 9800 and 93 respectively
Chen et al [53] (1) Autocorrelationfeatures Ultrasound 1020 (1)The obtained ROC area is 09840 plusmn 00072
10 Computational and Mathematical Methods in Medicine
Table 7 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Chen et al [61]
(1) Variance Contrast of WaveletCoefficient Ultrasound 242 (1)The achieved ROC curve 09396 plusmn 00183(2) Autocorrelation of WaveletCoefficient
Silva et al [62](1) 22 different morphologicalfeatures such as convexity andlobulation have been utilized
Ultrasound mdash (1)The best obtained Accuracy and ROCcurve are 9698 and 098 respectively
Saritas [63](1) Age of patient (2)massshape (3)mass border (4)Massdensity (5) BIRADS Mammogram mdash
(1) Disease prediction rate is 905(2) Neural Network utilized 5 neurons ininput layers and one hidden layer
Lopez-Melendez etal [64]
(1) Area perimeter etc havebeen utilized Mammogram 322 (1)The achieved Sensitivity and Specificity
are 9629 and 9900 respectively
themodel A kernel of size119898times119899 is scanned through the inputdata for the convolutional operation which ensures the localconnectivity and weight sharing property
(ii) Stride and Padding In the convolutional operation afilter scans through the input matrices In each step howmuch position a kernel filter moves through the matrixis known as the stride By default stride keeps to 1 Withinappropriate selection of the stride the model can lose theborder information To overcome this issue themodel utilizesextra rows and columns at the end of the matrices and theseadded rows and columns contain all 0s This adding of extrarows and columns which contain only zero value is known aszero padding
(iii) Nonlinear Operation The output of each of the kerneloperations is passed through a rectifier function such as Rec-tified Linear Unit (ReLU) Leaky-ReLU TanH and SigmoidThe Sigmoid function can be defined as
120590 (119909) = 1(1 + expminus119909) (3)
and the tanh function can be defined as
tanh (119909) = (exp119909 minus expminus119909)(exp119909 + expminus119909) (4)
However the most effective rectifier is ReLU The ReLUmethod converts all the information into zero if it is less thanor equal to zero and passes all the other data as is shown inFigure 13
120590 (119909) = max (0 119909) (5)
Another important nonlinear function is Leaky-RelU
where 120572 is predetermined parameter which can be varied togive a better model
minus3 minus2 minus1 0 1 2 3
1
2
3
InputO
utpu
t
Figure 13 ReLU Operation
(iv) Subsampling Subsampling is the procedure of reducingthe dimensionality of each of the feature maps of a particularlayer this operation is also known as a pooling operationActually it reduces the amount of feature information fromthe overall data By doing so it reduces the overall computa-tional complexity of themodel To do this 119904times119904 patch units areutilized The two most popular pooling methods are
(a) Max-Pooling
(b) Average Pooling
In Max-Pooling only the maximum values within a partic-ular kernel size are selected for further calculation Consideran example of a 16 times 16 image as shown in Figure 14 A 2 by2 kernel is applied to the whole image 4 blocks in total andproduces a 4 times 4 output image For each block of four valueswe have selected the maximum For instance from blocksone two three and four maximum values 4 40 13 and 8are selected respectively as they are the maximum in thatblock For the Average Pooling operation each kernel givesthe output as average
(v) Dropout Regularization of the weight can reduce theoutfitting problem Randomly removing some neurons can
Computational and Mathematical Methods in Medicine 11
Figure 15 Work-flow of a Convolutional Neural Network
regularize the overfilling problem The technique of ran-domly removing neurons from the network is known asdropout
(vi) Soft-Max Layer This layer contains normalized expo-nential functions to calculate the loss function for the dataclassification
Figure 15 shows a generalized CNN model for the imageclassificationAll the neurons of themost immediate layer of afully connected layer are completely connected with the fullyconnected layer like a conventional Neural Network Let119891119897minus1119895represent the 119895th feature map at the layer 119897minus1The 119895th featuremap at the layer 119897 can be represented as
where119873119897minus119897 represents the number of featuremaps at the 119897minus1thlayer 119896119894119895 represents the kernel function and 119887119897119895 represents thebias at 119897 where 120590 performs a nonlinear function operationThe layer before the Soft-Max Layer can be represented as
Let 119901 = 1 represent Benign class and 119901 = 2 represent theMalignant class The cross-entropy loss of the above functioncan be calculated as
119871119901 = minus ln (119910119901) (10)
Whichever group experiences a large loss value themodel will consider the other group as predicted class
A difficult part of working on DNN is that it requiresa specialized software package for the data analysis Fewresearch groups have been working on how effectively datacan be analyzed by DNN from different perspectives and thedemand Table 8 summarizes some of the software which isavailable for DNN analysis
The history of the CNN and its use for biomedical imageanalysis is a long one Fukushima first introduced a CNNnamed ldquonecognitronrdquo which has the ability to recognizestimulus patterns with a few shifting variances [113] Tothe best of our knowledge Wu et al first classified a setof mammogram images into malignant and benign classesusing a CNN model [78] In their proposed model they onlyutilized one hidden layer After that in 1996 Sahiner et alutilized CNNmodel to classify mass and normal breast tissueand achieved ROC scores of 087 [79] In 2002 Lo et alutilized aMultiple Circular Path CNN (MCPCNN) for tumoridentification from mammogram images and obtained ROCscores of around 089 After an absence of investigation ofthe CNN model this model regained its momentum afterthe work of Krizhevsky et al [114] Their proposed model isknown as AlexNet After this work a revolutionary change
12 Computational and Mathematical Methods in Medicine
Table 8 Available software for deep learning analysis
Software Interface and backend Provider
Caffe [65 66] Python MATLAB C++ Berkeley Vision and Learning CentreUniversity of California Berkeley
Torch [67] C LuaJIT
MatConvNet [68 69] MATLAB C Visual Geometry Group Department ofEngineering University of Oxford
Theano [70 71] Python Montreal Institute for Learning AlgorithmsUniversity of Montreal
TensorFlows [72] C++ Python GoogleCNTK [73] C++ MicrosoftKeras [74] Theano Tensor Flow MITdl4j [75] Java Skymind Engineering
DeeBNET [76 77] MATLAB Information Technology DepartmentAmirkabir University of Technology
has been achieved in the image classification and analysisfield As an advanced engineering of the AlexNet the papertitled ldquoGoing Deeper with Convolutionsrdquo by Szegedy [115]introduced the GoogleNet model This model contains amuch deeper network than AlexNet Sequentially ResNet[116] Inception [117] Inception-v4 Inception-ResNet [118]and a few other models have recently been introduced
Later directly or with some advanced modificationthese DNN models have been adapted for biomedical imageanalysis In 2015 Fonseca et al [81] classified breast densityusing CNN techniques CNN requires a sufficient amountof data to train the system It is always very difficult tofind a sufficient amount of medical data for training a CNNmodel A pretrained CNN model with some fine tuning canbe used rather than create a model from scratch [119] Theauthors of [119] did not perform their experiments on a breastcancer image dataset however they have performed theirexperiments on three different medical datasets with layer-wise training and claimed that ldquoretrained CNN along withadequate training can provide better or at least the sameamount of performancerdquo
The Deep Belief Network (DBN) is another branch of theDeep Neural Network which mainly consists of RestrictedBoltzmann Machine (RBM) techniques The DBN methodwas first utilized for supervised image classification by Liu etal [120] After that Abdel-Zaher and Eldeib utilized the DBNmethod for breast image classification [121] This field is stillnot fully explored for breast image classification yet Zhanget al utilized both RBM and Point-Wise Gated RBM (PRBM)for shear-wave electrography image classification where thedataset contains 227 images [97]Their achieved classificationAccuracy Sensitivity and Specificity are 9340 8860 and9710 respectively Tables 9 10 and 11 have summarized themost recent work for breast image classification along withsome pioneer work on CNN
313 Logic Based Algorithm A Logic Based algorithm isa very popular and effective classification method whichfollows the tree structure principle and logical argument asshown in Figure 16 This algorithm classifies instances based
on the featurersquos values Along with other criteria a decision-tree based algorithm contains the following features
(i) Root node a root node contains no incoming nodeand it may or may not contain any outgoing edge
(ii) Splitting splitting is the process of subdividing a set ofcases into a particular group Normally the followingcriteria are maintained for the splitting
(a) information gain(b) Gini index(c) chi squared
(iii) Decision node(iv) Leafterminal node this kind of node has exactly one
incoming edge and no outgoing edgeThe tree alwaysterminates here with a decision
(v) Pruning pruning is a process of removing subtreesfrom the tree Pruning performs to reduce the over-fitting problem Two kinds of pruning techniques areavailable
(a) prepruning(b) postpruning
Among all the tree based algorithms IterativeDichotomiser 3 (ID3) can be considered as a pioneerproposed by Quinlan [149] The problem of the ID3algorithm is to find the optimal solution which is very muchprone towards overfitting To overcome the limitation of theID3 algorithm the C45 algorithm has been introduced byQuinlan [150] where a pruning method has been introducedto control the overfitting problem Pritom et al [151] classifiedthe Wisconsin breast dataset where they utilized 35 featuresThey have obtained 7630 Accuracy 7510 False PositiveRate and ROC score 0745 when they ranked the featuresWithout ranking the features they obtained 7370Accuracy5070 False Positive Rate and ROC score value 5280 Asriet al [152] utilized the C45 algorithm for the Wisconsin
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Figure 7 Classification of features for breast image classification
of the image Breast Imaging-Reporting and Data System(BI-RADS) is a mammography image assessment techniquecontaining 6 categories normally assigned by the radiologistFeature detector actually provides information whether theparticular feature is available in the image or not Structuralfeatures provide information about the features structure andorientation such as the area Convex Hull and centroid Thiskind of information gives more detailed information aboutthe features In a cancer image it can provide the area ofthe nucleus or the centroid of the mass Mean Medianand Standard Deviation always provide some importantinformation on the dataset and their distribution This kindof features has been categorized as statistical features Thetotal hierarchy of the image feature extraction is resented inFigure 7 Tables 2 and 3 further summarize the local featuresin detail
Features which are extracted for classification do notalways carry the same importance Some features may evencontribute to degrading the classifier performance Priori-tization of the feature set can reduce the classifier modelcomplexity and so it can reduce the computational timeFeature set selection and prioritization can be classified intothree broad categories
(i) Filter the filter method selects features without eval-uating any classifier algorithm
(ii) Wrapper the wrapper method selects the feature setbased on the evaluation performance of a particularclassifier
(iii) Embedded the embeddedmethod takes advantage ofthe filter andwrappermethods for classifier construc-tion
6 Computational and Mathematical Methods in Medicine
Table 3 Feature descriptor
Feature category Feature descriptionStatistical (1)Mean (2)Median (3) Standard Deviation (4) Skewness (5) Kurtosis (6) Range
Descriptor
(1) Scale Invariant Feature Transform (SIFT) [17 19] (2) Gradient Location-Orientation Histogram (GLOH) [20] (3)Speeded-Up Robust Features Descriptor (SURF) [21ndash23] (4) Local Binary Pattern (LBP) [24ndash27] (5) Binary RobustIndependent Elementary Features (BRIEF) [28] (6)Weber Local Descriptor (WLD) [29 30] (7) Back Ground LocalBinary Pattern (BGLBP) [31] (8) Center-Symmetric Local Binary Pattern (CS-LBP) [32] (9) Second-OrderCenter-Symmetric Local Derivative Pattern (CS-LBP) [33] (10) Center-Symmetric Scale Invariant Local TernaryPatterns (CS-SILTP) [34] (11) Extended LBP or Circular LBP (E-LBP) [35] (12)Opponent Color Local Binary Pattern(OC-LBP) [36] (13) Original LBP(O-LBP) [25] (14) Spatial Extended Center-Symmetric Local Binary Pattern(SCS-LBP) [37] (15) Scale Invariant Local Ternary Pattern (SI-LTP) [38] (16) Variance-Based LBP (VAR-LBP) [24](17) eXtended Center-Symmetric Local Binary Pattern (XCS-LBP) (18) Average Local Binary Pattern (ALBP) (19)Block Based Local Binary Pattern (BBLBP) [39]
Figure 8 shows a generalized feature selection methodwhere we have further classified the filter method intoFisher Score Mutual Information Relief and chi squaremethods The embedded method has been classified intoBridge Regularization Lasso and Adaptive Lasso methodswhile the wrapper method has been classified to recursivefeature selection and sequential feature selection method
23 Classifier Model Based on the learning point of viewbreast image classification techniques can be categorized intothe following three classes [41]
These three classes can be split into Deep Neural Network(DNN) and conventional classifier (without DNN) and tosome further classes as in Table 4
24 Performance Measuring Parameter A Confusion Matrixis a two-dimensional table which is used to a give a visual
True
clas
s
Hypothesized class
True positive (A) False negative (B)
False positive (C) True negative (D)
Figure 9 Confusion Matrix
perception of classification experiments [54] The (119894 119895)thposition of the confusion table indicates the number of timesthat the 119894th object is classified as the 119895th object The diagonalof this matrix indicates the number of times the objects arecorrectly classified Figure 9 shows a graphical representationof a Confusion Matrix for the binary classification case
Computational and Mathematical Methods in Medicine 7
Table 4 A simplified hierarchy of classification
Learning technique Algorithm
Supervised
Conventional
(a) Logic based
(1) ID3 (2) C45 (3) bagging(4) random trees (5) Random Forest(6) boosting (7) advanced boosting(8) Extreme Boosting (XGBoosting)
(a) Self-training(b) Graph Based(c) S3V3(d) Multiview(e) Generative model
Among the different classification performance proper-ties this matrix will provide following parameters
(i) Recall is defined as Recall = TP(TP + FN)(ii) Precision is defined as Precision = TP(TP + FP)(iii) Specificity is defined as Specificity = TN(TN + FP)(iv) Accuracy is defined as ACC = (TP+TN)(TP+TN+
FP + FN)(v) F-1 score is defined as 1198651 = (2 times Recall)(2 times Recall +
FP + FN)(vi) Matthew Correlation Coefficient (MCC) MCC is a
performance parameter of a binary classifier in therange minus1 to +1 If the MCC values trend moretowards +1 the classifier gives a more accurate classi-fier and the opposite condition will occur if the valueof theMCC trend towards theminus1MCCcanbe definedas
MCC
= TP times TN minus FP times FNradic(TP + FP) (TP + FN) (TN + FP) (TN + FP) (1)
3 Performance of Different Classifier Modelon Breast Images Dataset
Based on Supervised Semisupervised and Unsupervisedmethods different research groups have been performedclassification operation on different image database In thissection we have summarized few of the works of breast imageclassification
31 Performance Based on Supervised Learning In super-vised learning a general hypothesis is established based onexternally supplied instances to produce future predictionFor the supervised classification task features are extractedor automatically crafted from the available dataset and eachsample is mapped to a dedicated class With the help of thefeatures and their levels a hypothesis is created Based on thehypothesis unknown data are classified [55]
Figure 10 represents an overall supervised classifier archi-tecture In general the whole dataset is split into trainingand testing parts To validate the data some time dataare also split into a validation part as well After the datasplitting themost important part is to find out the appropriatefeatures to classify the data with the utmost AccuracyFinding the features can be classified into two categorieslocally and globally crafted Locally crafted means that thismethod requires a hand-held exercise to find out the featureswhereas globally craftedmeans that a kernelmethod has beenintroduced for the feature extraction Handcrafted featurescan be prioritized whereas Global Feature selection does nothave this luxury
311 Conventional Neural Network The Neural Network(NN) concept comes from the working principle of thehuman brain A biological neuron consists of the followingfour parts
8 Computational and Mathematical Methods in Medicine
Classifier model
Imagedatabase
Traintestdata splitting Locally
craftedGloballycrafted
Hand crafting
Kernel basedcrafting
Featureprioritization
Conventionalclassifier
DNNclassifier
Evaluationmatrix
Classifieddata
Feature collection
Ensemble learning
Figure 10 A generalized supervised classifier model
Nucleus
Axon
Cell body
Dendrites
Figure 11 A model of a biological neuron
Dendrites collect signals and axons carry the signal to thenext dendrite after processing by the cell body as shown inFigure 11 Using the neuronworking principle the perceptronmodel was proposed by Rosenblatt in 1957 [56] A single-layer perceptron linearly combines the input signal and givesa decision based on a threshold function Based on theworking principle and with some advanced mechanism andengineering NNmethods have established a strong footprintin many problem-solving issues Figure 12 shows the basicworking principle of NN techniques
In the NN model the input data X = 1199090 1199091 119909119873 isfirst multiplied by the weight dataW = 1199080 1199081 119908119873 andthen the output is calculated using
Y = g (sum) wheresum = W sdot X (2)
Function g is known as the activation function Thisfunction can be any threshold value or Sigmoid or hyperbolicand so forth In the early stages feed-forwardNeuralNetworktechniques were introduced [57] lately the backpropagationmethod has been invented to utilize the error information toimprove the system performance [58 59]
The history of breast image classification by NN is a longone To the best of my knowledge a lot of the pioneer work
yg
x0
x1
xNminus1
xN
w0
w1
wNminus1
wN
Figure 12Working principle of a simpleNeuralNetwork technique
was performed by Dawson et al in 1991 [60] Since then NNhas been utilized as one of the strong tools for breast imageclassification We have summarized some of the work relatedto NN and breast image classification in Tables 5 6 and 7
312 Deep Neural Network Deep Neural Network (DNN) isa state-of-the-art concept where conventional NN techniqueshave been utilized with advanced engineering It is foundthat conventional NNs have difficulties in solving complexproblems whereas DNNs solve them with utmost PrecisionHowever DNNs suffer from more time and computationalcomplexity than the conventional NN
Convolutional Neural Network A CNN model is the combi-nation of a few intermediate mathematical structures Thisintermediatemathematical structure creates or helps to createdifferent layers
(i) Convolutional Layer Among all the other layers theconvolutional layer is considered as the most important partfor a CNN model and can be considered as the backbone of
Computational and Mathematical Methods in Medicine 9
Table 5 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Rajakeerthana et al [42] (1) GLCM GLDM SRDMNGLCM GLRM Mammogram 322 (1)The classifier achieved 9920
Accuracy
Lessa and Marengoni [43](1)Mean Median StandardDeviation Skewness KurtosisEntropy Range
Wan et al [44] (1) ALBP (2) BBLBP OCM 46(1) Achieved Sensitivity and Specificityare 100 and 8520 respectively(2) ROC value obtained 0959
Chen et al [40] (1) 19 BI-RADS features havebeen used Ultrasound 238
(1) Chi squared method has beenutilized for the feature selection(2) Achieved Accuracy Sensitivity andSpecificity are 9610 9670 and9570 respectively
de Lima et al [45] (1) Total 416 features have beenused Mammogram 355
(1)Multiresolution wavelet and Zernikemoment have been utilized for thefeature extraction
Abirami et al [46](1) 12 statistical measures such asMean Median and Max havebeen utilized as the features
Mammogram 322
(1)Wavelet transform has been utilizedfor the feature extraction(2)The achieved Accuracy Sensitivityand Specificity are 9550 9500 and9600 respectively
El Atlas et al [47] (1) 13 morphological featureshave been utilized Mammogram 410
(1) Firstly the edge information hasbeen utilized for the mass segmentationand then the morphological featureswere extracted(2) Achieved best Accuracy is 975
Table 6 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Alharbi et al [48] (1) 49 features havebeen utilized Mammogram 1100
(1) Five feature selection methods Fisher scoreMinimum Redundancy-Maximum Relevance Relief-fSequential Forward Feature Selection and GeneticAlgorithm have been used(2) Achieved Accuracy Sensitivity and specificity are9420 9836 and 9927 respectively
Peng et al [49](1)Haralick andTamura features havebeen utilized
Mammogram 322
(1) Feature reduction has been performed byRough-Set theory and selected 5 prioritized features(2)The best Accuracy Sensitivity and Specificityachieved were 9600 9860 and 8930
Jalalian et al [50] (1) GLCM Mammogram(1)The obtained classifier Accuracy Sensitivity andSpecificity are 9520 9240 and 9800respectively(2) Compactness
Li et al [51](1) Four featurevectors have beencalculated
Mammogram 322
(1) 2D contour of breast mass in mammography hasbeen converted into 1D signature(2) NN techniques achieved Accuracy is 9960 whenRMS slope is utilized
Chen et al [52] (1) Autocorrelationfeatures Ultrasound 242 (1)The overall achieved Accuracy Sensitivity and
Specificity are 9500 9800 and 93 respectively
Chen et al [53] (1) Autocorrelationfeatures Ultrasound 1020 (1)The obtained ROC area is 09840 plusmn 00072
10 Computational and Mathematical Methods in Medicine
Table 7 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Chen et al [61]
(1) Variance Contrast of WaveletCoefficient Ultrasound 242 (1)The achieved ROC curve 09396 plusmn 00183(2) Autocorrelation of WaveletCoefficient
Silva et al [62](1) 22 different morphologicalfeatures such as convexity andlobulation have been utilized
Ultrasound mdash (1)The best obtained Accuracy and ROCcurve are 9698 and 098 respectively
Saritas [63](1) Age of patient (2)massshape (3)mass border (4)Massdensity (5) BIRADS Mammogram mdash
(1) Disease prediction rate is 905(2) Neural Network utilized 5 neurons ininput layers and one hidden layer
Lopez-Melendez etal [64]
(1) Area perimeter etc havebeen utilized Mammogram 322 (1)The achieved Sensitivity and Specificity
are 9629 and 9900 respectively
themodel A kernel of size119898times119899 is scanned through the inputdata for the convolutional operation which ensures the localconnectivity and weight sharing property
(ii) Stride and Padding In the convolutional operation afilter scans through the input matrices In each step howmuch position a kernel filter moves through the matrixis known as the stride By default stride keeps to 1 Withinappropriate selection of the stride the model can lose theborder information To overcome this issue themodel utilizesextra rows and columns at the end of the matrices and theseadded rows and columns contain all 0s This adding of extrarows and columns which contain only zero value is known aszero padding
(iii) Nonlinear Operation The output of each of the kerneloperations is passed through a rectifier function such as Rec-tified Linear Unit (ReLU) Leaky-ReLU TanH and SigmoidThe Sigmoid function can be defined as
120590 (119909) = 1(1 + expminus119909) (3)
and the tanh function can be defined as
tanh (119909) = (exp119909 minus expminus119909)(exp119909 + expminus119909) (4)
However the most effective rectifier is ReLU The ReLUmethod converts all the information into zero if it is less thanor equal to zero and passes all the other data as is shown inFigure 13
120590 (119909) = max (0 119909) (5)
Another important nonlinear function is Leaky-RelU
where 120572 is predetermined parameter which can be varied togive a better model
minus3 minus2 minus1 0 1 2 3
1
2
3
InputO
utpu
t
Figure 13 ReLU Operation
(iv) Subsampling Subsampling is the procedure of reducingthe dimensionality of each of the feature maps of a particularlayer this operation is also known as a pooling operationActually it reduces the amount of feature information fromthe overall data By doing so it reduces the overall computa-tional complexity of themodel To do this 119904times119904 patch units areutilized The two most popular pooling methods are
(a) Max-Pooling
(b) Average Pooling
In Max-Pooling only the maximum values within a partic-ular kernel size are selected for further calculation Consideran example of a 16 times 16 image as shown in Figure 14 A 2 by2 kernel is applied to the whole image 4 blocks in total andproduces a 4 times 4 output image For each block of four valueswe have selected the maximum For instance from blocksone two three and four maximum values 4 40 13 and 8are selected respectively as they are the maximum in thatblock For the Average Pooling operation each kernel givesthe output as average
(v) Dropout Regularization of the weight can reduce theoutfitting problem Randomly removing some neurons can
Computational and Mathematical Methods in Medicine 11
Figure 15 Work-flow of a Convolutional Neural Network
regularize the overfilling problem The technique of ran-domly removing neurons from the network is known asdropout
(vi) Soft-Max Layer This layer contains normalized expo-nential functions to calculate the loss function for the dataclassification
Figure 15 shows a generalized CNN model for the imageclassificationAll the neurons of themost immediate layer of afully connected layer are completely connected with the fullyconnected layer like a conventional Neural Network Let119891119897minus1119895represent the 119895th feature map at the layer 119897minus1The 119895th featuremap at the layer 119897 can be represented as
where119873119897minus119897 represents the number of featuremaps at the 119897minus1thlayer 119896119894119895 represents the kernel function and 119887119897119895 represents thebias at 119897 where 120590 performs a nonlinear function operationThe layer before the Soft-Max Layer can be represented as
Let 119901 = 1 represent Benign class and 119901 = 2 represent theMalignant class The cross-entropy loss of the above functioncan be calculated as
119871119901 = minus ln (119910119901) (10)
Whichever group experiences a large loss value themodel will consider the other group as predicted class
A difficult part of working on DNN is that it requiresa specialized software package for the data analysis Fewresearch groups have been working on how effectively datacan be analyzed by DNN from different perspectives and thedemand Table 8 summarizes some of the software which isavailable for DNN analysis
The history of the CNN and its use for biomedical imageanalysis is a long one Fukushima first introduced a CNNnamed ldquonecognitronrdquo which has the ability to recognizestimulus patterns with a few shifting variances [113] Tothe best of our knowledge Wu et al first classified a setof mammogram images into malignant and benign classesusing a CNN model [78] In their proposed model they onlyutilized one hidden layer After that in 1996 Sahiner et alutilized CNNmodel to classify mass and normal breast tissueand achieved ROC scores of 087 [79] In 2002 Lo et alutilized aMultiple Circular Path CNN (MCPCNN) for tumoridentification from mammogram images and obtained ROCscores of around 089 After an absence of investigation ofthe CNN model this model regained its momentum afterthe work of Krizhevsky et al [114] Their proposed model isknown as AlexNet After this work a revolutionary change
12 Computational and Mathematical Methods in Medicine
Table 8 Available software for deep learning analysis
Software Interface and backend Provider
Caffe [65 66] Python MATLAB C++ Berkeley Vision and Learning CentreUniversity of California Berkeley
Torch [67] C LuaJIT
MatConvNet [68 69] MATLAB C Visual Geometry Group Department ofEngineering University of Oxford
Theano [70 71] Python Montreal Institute for Learning AlgorithmsUniversity of Montreal
TensorFlows [72] C++ Python GoogleCNTK [73] C++ MicrosoftKeras [74] Theano Tensor Flow MITdl4j [75] Java Skymind Engineering
DeeBNET [76 77] MATLAB Information Technology DepartmentAmirkabir University of Technology
has been achieved in the image classification and analysisfield As an advanced engineering of the AlexNet the papertitled ldquoGoing Deeper with Convolutionsrdquo by Szegedy [115]introduced the GoogleNet model This model contains amuch deeper network than AlexNet Sequentially ResNet[116] Inception [117] Inception-v4 Inception-ResNet [118]and a few other models have recently been introduced
Later directly or with some advanced modificationthese DNN models have been adapted for biomedical imageanalysis In 2015 Fonseca et al [81] classified breast densityusing CNN techniques CNN requires a sufficient amountof data to train the system It is always very difficult tofind a sufficient amount of medical data for training a CNNmodel A pretrained CNN model with some fine tuning canbe used rather than create a model from scratch [119] Theauthors of [119] did not perform their experiments on a breastcancer image dataset however they have performed theirexperiments on three different medical datasets with layer-wise training and claimed that ldquoretrained CNN along withadequate training can provide better or at least the sameamount of performancerdquo
The Deep Belief Network (DBN) is another branch of theDeep Neural Network which mainly consists of RestrictedBoltzmann Machine (RBM) techniques The DBN methodwas first utilized for supervised image classification by Liu etal [120] After that Abdel-Zaher and Eldeib utilized the DBNmethod for breast image classification [121] This field is stillnot fully explored for breast image classification yet Zhanget al utilized both RBM and Point-Wise Gated RBM (PRBM)for shear-wave electrography image classification where thedataset contains 227 images [97]Their achieved classificationAccuracy Sensitivity and Specificity are 9340 8860 and9710 respectively Tables 9 10 and 11 have summarized themost recent work for breast image classification along withsome pioneer work on CNN
313 Logic Based Algorithm A Logic Based algorithm isa very popular and effective classification method whichfollows the tree structure principle and logical argument asshown in Figure 16 This algorithm classifies instances based
on the featurersquos values Along with other criteria a decision-tree based algorithm contains the following features
(i) Root node a root node contains no incoming nodeand it may or may not contain any outgoing edge
(ii) Splitting splitting is the process of subdividing a set ofcases into a particular group Normally the followingcriteria are maintained for the splitting
(a) information gain(b) Gini index(c) chi squared
(iii) Decision node(iv) Leafterminal node this kind of node has exactly one
incoming edge and no outgoing edgeThe tree alwaysterminates here with a decision
(v) Pruning pruning is a process of removing subtreesfrom the tree Pruning performs to reduce the over-fitting problem Two kinds of pruning techniques areavailable
(a) prepruning(b) postpruning
Among all the tree based algorithms IterativeDichotomiser 3 (ID3) can be considered as a pioneerproposed by Quinlan [149] The problem of the ID3algorithm is to find the optimal solution which is very muchprone towards overfitting To overcome the limitation of theID3 algorithm the C45 algorithm has been introduced byQuinlan [150] where a pruning method has been introducedto control the overfitting problem Pritom et al [151] classifiedthe Wisconsin breast dataset where they utilized 35 featuresThey have obtained 7630 Accuracy 7510 False PositiveRate and ROC score 0745 when they ranked the featuresWithout ranking the features they obtained 7370Accuracy5070 False Positive Rate and ROC score value 5280 Asriet al [152] utilized the C45 algorithm for the Wisconsin
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Figure 8 shows a generalized feature selection methodwhere we have further classified the filter method intoFisher Score Mutual Information Relief and chi squaremethods The embedded method has been classified intoBridge Regularization Lasso and Adaptive Lasso methodswhile the wrapper method has been classified to recursivefeature selection and sequential feature selection method
23 Classifier Model Based on the learning point of viewbreast image classification techniques can be categorized intothe following three classes [41]
These three classes can be split into Deep Neural Network(DNN) and conventional classifier (without DNN) and tosome further classes as in Table 4
24 Performance Measuring Parameter A Confusion Matrixis a two-dimensional table which is used to a give a visual
True
clas
s
Hypothesized class
True positive (A) False negative (B)
False positive (C) True negative (D)
Figure 9 Confusion Matrix
perception of classification experiments [54] The (119894 119895)thposition of the confusion table indicates the number of timesthat the 119894th object is classified as the 119895th object The diagonalof this matrix indicates the number of times the objects arecorrectly classified Figure 9 shows a graphical representationof a Confusion Matrix for the binary classification case
Computational and Mathematical Methods in Medicine 7
Table 4 A simplified hierarchy of classification
Learning technique Algorithm
Supervised
Conventional
(a) Logic based
(1) ID3 (2) C45 (3) bagging(4) random trees (5) Random Forest(6) boosting (7) advanced boosting(8) Extreme Boosting (XGBoosting)
(a) Self-training(b) Graph Based(c) S3V3(d) Multiview(e) Generative model
Among the different classification performance proper-ties this matrix will provide following parameters
(i) Recall is defined as Recall = TP(TP + FN)(ii) Precision is defined as Precision = TP(TP + FP)(iii) Specificity is defined as Specificity = TN(TN + FP)(iv) Accuracy is defined as ACC = (TP+TN)(TP+TN+
FP + FN)(v) F-1 score is defined as 1198651 = (2 times Recall)(2 times Recall +
FP + FN)(vi) Matthew Correlation Coefficient (MCC) MCC is a
performance parameter of a binary classifier in therange minus1 to +1 If the MCC values trend moretowards +1 the classifier gives a more accurate classi-fier and the opposite condition will occur if the valueof theMCC trend towards theminus1MCCcanbe definedas
MCC
= TP times TN minus FP times FNradic(TP + FP) (TP + FN) (TN + FP) (TN + FP) (1)
3 Performance of Different Classifier Modelon Breast Images Dataset
Based on Supervised Semisupervised and Unsupervisedmethods different research groups have been performedclassification operation on different image database In thissection we have summarized few of the works of breast imageclassification
31 Performance Based on Supervised Learning In super-vised learning a general hypothesis is established based onexternally supplied instances to produce future predictionFor the supervised classification task features are extractedor automatically crafted from the available dataset and eachsample is mapped to a dedicated class With the help of thefeatures and their levels a hypothesis is created Based on thehypothesis unknown data are classified [55]
Figure 10 represents an overall supervised classifier archi-tecture In general the whole dataset is split into trainingand testing parts To validate the data some time dataare also split into a validation part as well After the datasplitting themost important part is to find out the appropriatefeatures to classify the data with the utmost AccuracyFinding the features can be classified into two categorieslocally and globally crafted Locally crafted means that thismethod requires a hand-held exercise to find out the featureswhereas globally craftedmeans that a kernelmethod has beenintroduced for the feature extraction Handcrafted featurescan be prioritized whereas Global Feature selection does nothave this luxury
311 Conventional Neural Network The Neural Network(NN) concept comes from the working principle of thehuman brain A biological neuron consists of the followingfour parts
8 Computational and Mathematical Methods in Medicine
Classifier model
Imagedatabase
Traintestdata splitting Locally
craftedGloballycrafted
Hand crafting
Kernel basedcrafting
Featureprioritization
Conventionalclassifier
DNNclassifier
Evaluationmatrix
Classifieddata
Feature collection
Ensemble learning
Figure 10 A generalized supervised classifier model
Nucleus
Axon
Cell body
Dendrites
Figure 11 A model of a biological neuron
Dendrites collect signals and axons carry the signal to thenext dendrite after processing by the cell body as shown inFigure 11 Using the neuronworking principle the perceptronmodel was proposed by Rosenblatt in 1957 [56] A single-layer perceptron linearly combines the input signal and givesa decision based on a threshold function Based on theworking principle and with some advanced mechanism andengineering NNmethods have established a strong footprintin many problem-solving issues Figure 12 shows the basicworking principle of NN techniques
In the NN model the input data X = 1199090 1199091 119909119873 isfirst multiplied by the weight dataW = 1199080 1199081 119908119873 andthen the output is calculated using
Y = g (sum) wheresum = W sdot X (2)
Function g is known as the activation function Thisfunction can be any threshold value or Sigmoid or hyperbolicand so forth In the early stages feed-forwardNeuralNetworktechniques were introduced [57] lately the backpropagationmethod has been invented to utilize the error information toimprove the system performance [58 59]
The history of breast image classification by NN is a longone To the best of my knowledge a lot of the pioneer work
yg
x0
x1
xNminus1
xN
w0
w1
wNminus1
wN
Figure 12Working principle of a simpleNeuralNetwork technique
was performed by Dawson et al in 1991 [60] Since then NNhas been utilized as one of the strong tools for breast imageclassification We have summarized some of the work relatedto NN and breast image classification in Tables 5 6 and 7
312 Deep Neural Network Deep Neural Network (DNN) isa state-of-the-art concept where conventional NN techniqueshave been utilized with advanced engineering It is foundthat conventional NNs have difficulties in solving complexproblems whereas DNNs solve them with utmost PrecisionHowever DNNs suffer from more time and computationalcomplexity than the conventional NN
Convolutional Neural Network A CNN model is the combi-nation of a few intermediate mathematical structures Thisintermediatemathematical structure creates or helps to createdifferent layers
(i) Convolutional Layer Among all the other layers theconvolutional layer is considered as the most important partfor a CNN model and can be considered as the backbone of
Computational and Mathematical Methods in Medicine 9
Table 5 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Rajakeerthana et al [42] (1) GLCM GLDM SRDMNGLCM GLRM Mammogram 322 (1)The classifier achieved 9920
Accuracy
Lessa and Marengoni [43](1)Mean Median StandardDeviation Skewness KurtosisEntropy Range
Wan et al [44] (1) ALBP (2) BBLBP OCM 46(1) Achieved Sensitivity and Specificityare 100 and 8520 respectively(2) ROC value obtained 0959
Chen et al [40] (1) 19 BI-RADS features havebeen used Ultrasound 238
(1) Chi squared method has beenutilized for the feature selection(2) Achieved Accuracy Sensitivity andSpecificity are 9610 9670 and9570 respectively
de Lima et al [45] (1) Total 416 features have beenused Mammogram 355
(1)Multiresolution wavelet and Zernikemoment have been utilized for thefeature extraction
Abirami et al [46](1) 12 statistical measures such asMean Median and Max havebeen utilized as the features
Mammogram 322
(1)Wavelet transform has been utilizedfor the feature extraction(2)The achieved Accuracy Sensitivityand Specificity are 9550 9500 and9600 respectively
El Atlas et al [47] (1) 13 morphological featureshave been utilized Mammogram 410
(1) Firstly the edge information hasbeen utilized for the mass segmentationand then the morphological featureswere extracted(2) Achieved best Accuracy is 975
Table 6 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Alharbi et al [48] (1) 49 features havebeen utilized Mammogram 1100
(1) Five feature selection methods Fisher scoreMinimum Redundancy-Maximum Relevance Relief-fSequential Forward Feature Selection and GeneticAlgorithm have been used(2) Achieved Accuracy Sensitivity and specificity are9420 9836 and 9927 respectively
Peng et al [49](1)Haralick andTamura features havebeen utilized
Mammogram 322
(1) Feature reduction has been performed byRough-Set theory and selected 5 prioritized features(2)The best Accuracy Sensitivity and Specificityachieved were 9600 9860 and 8930
Jalalian et al [50] (1) GLCM Mammogram(1)The obtained classifier Accuracy Sensitivity andSpecificity are 9520 9240 and 9800respectively(2) Compactness
Li et al [51](1) Four featurevectors have beencalculated
Mammogram 322
(1) 2D contour of breast mass in mammography hasbeen converted into 1D signature(2) NN techniques achieved Accuracy is 9960 whenRMS slope is utilized
Chen et al [52] (1) Autocorrelationfeatures Ultrasound 242 (1)The overall achieved Accuracy Sensitivity and
Specificity are 9500 9800 and 93 respectively
Chen et al [53] (1) Autocorrelationfeatures Ultrasound 1020 (1)The obtained ROC area is 09840 plusmn 00072
10 Computational and Mathematical Methods in Medicine
Table 7 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Chen et al [61]
(1) Variance Contrast of WaveletCoefficient Ultrasound 242 (1)The achieved ROC curve 09396 plusmn 00183(2) Autocorrelation of WaveletCoefficient
Silva et al [62](1) 22 different morphologicalfeatures such as convexity andlobulation have been utilized
Ultrasound mdash (1)The best obtained Accuracy and ROCcurve are 9698 and 098 respectively
Saritas [63](1) Age of patient (2)massshape (3)mass border (4)Massdensity (5) BIRADS Mammogram mdash
(1) Disease prediction rate is 905(2) Neural Network utilized 5 neurons ininput layers and one hidden layer
Lopez-Melendez etal [64]
(1) Area perimeter etc havebeen utilized Mammogram 322 (1)The achieved Sensitivity and Specificity
are 9629 and 9900 respectively
themodel A kernel of size119898times119899 is scanned through the inputdata for the convolutional operation which ensures the localconnectivity and weight sharing property
(ii) Stride and Padding In the convolutional operation afilter scans through the input matrices In each step howmuch position a kernel filter moves through the matrixis known as the stride By default stride keeps to 1 Withinappropriate selection of the stride the model can lose theborder information To overcome this issue themodel utilizesextra rows and columns at the end of the matrices and theseadded rows and columns contain all 0s This adding of extrarows and columns which contain only zero value is known aszero padding
(iii) Nonlinear Operation The output of each of the kerneloperations is passed through a rectifier function such as Rec-tified Linear Unit (ReLU) Leaky-ReLU TanH and SigmoidThe Sigmoid function can be defined as
120590 (119909) = 1(1 + expminus119909) (3)
and the tanh function can be defined as
tanh (119909) = (exp119909 minus expminus119909)(exp119909 + expminus119909) (4)
However the most effective rectifier is ReLU The ReLUmethod converts all the information into zero if it is less thanor equal to zero and passes all the other data as is shown inFigure 13
120590 (119909) = max (0 119909) (5)
Another important nonlinear function is Leaky-RelU
where 120572 is predetermined parameter which can be varied togive a better model
minus3 minus2 minus1 0 1 2 3
1
2
3
InputO
utpu
t
Figure 13 ReLU Operation
(iv) Subsampling Subsampling is the procedure of reducingthe dimensionality of each of the feature maps of a particularlayer this operation is also known as a pooling operationActually it reduces the amount of feature information fromthe overall data By doing so it reduces the overall computa-tional complexity of themodel To do this 119904times119904 patch units areutilized The two most popular pooling methods are
(a) Max-Pooling
(b) Average Pooling
In Max-Pooling only the maximum values within a partic-ular kernel size are selected for further calculation Consideran example of a 16 times 16 image as shown in Figure 14 A 2 by2 kernel is applied to the whole image 4 blocks in total andproduces a 4 times 4 output image For each block of four valueswe have selected the maximum For instance from blocksone two three and four maximum values 4 40 13 and 8are selected respectively as they are the maximum in thatblock For the Average Pooling operation each kernel givesthe output as average
(v) Dropout Regularization of the weight can reduce theoutfitting problem Randomly removing some neurons can
Computational and Mathematical Methods in Medicine 11
Figure 15 Work-flow of a Convolutional Neural Network
regularize the overfilling problem The technique of ran-domly removing neurons from the network is known asdropout
(vi) Soft-Max Layer This layer contains normalized expo-nential functions to calculate the loss function for the dataclassification
Figure 15 shows a generalized CNN model for the imageclassificationAll the neurons of themost immediate layer of afully connected layer are completely connected with the fullyconnected layer like a conventional Neural Network Let119891119897minus1119895represent the 119895th feature map at the layer 119897minus1The 119895th featuremap at the layer 119897 can be represented as
where119873119897minus119897 represents the number of featuremaps at the 119897minus1thlayer 119896119894119895 represents the kernel function and 119887119897119895 represents thebias at 119897 where 120590 performs a nonlinear function operationThe layer before the Soft-Max Layer can be represented as
Let 119901 = 1 represent Benign class and 119901 = 2 represent theMalignant class The cross-entropy loss of the above functioncan be calculated as
119871119901 = minus ln (119910119901) (10)
Whichever group experiences a large loss value themodel will consider the other group as predicted class
A difficult part of working on DNN is that it requiresa specialized software package for the data analysis Fewresearch groups have been working on how effectively datacan be analyzed by DNN from different perspectives and thedemand Table 8 summarizes some of the software which isavailable for DNN analysis
The history of the CNN and its use for biomedical imageanalysis is a long one Fukushima first introduced a CNNnamed ldquonecognitronrdquo which has the ability to recognizestimulus patterns with a few shifting variances [113] Tothe best of our knowledge Wu et al first classified a setof mammogram images into malignant and benign classesusing a CNN model [78] In their proposed model they onlyutilized one hidden layer After that in 1996 Sahiner et alutilized CNNmodel to classify mass and normal breast tissueand achieved ROC scores of 087 [79] In 2002 Lo et alutilized aMultiple Circular Path CNN (MCPCNN) for tumoridentification from mammogram images and obtained ROCscores of around 089 After an absence of investigation ofthe CNN model this model regained its momentum afterthe work of Krizhevsky et al [114] Their proposed model isknown as AlexNet After this work a revolutionary change
12 Computational and Mathematical Methods in Medicine
Table 8 Available software for deep learning analysis
Software Interface and backend Provider
Caffe [65 66] Python MATLAB C++ Berkeley Vision and Learning CentreUniversity of California Berkeley
Torch [67] C LuaJIT
MatConvNet [68 69] MATLAB C Visual Geometry Group Department ofEngineering University of Oxford
Theano [70 71] Python Montreal Institute for Learning AlgorithmsUniversity of Montreal
TensorFlows [72] C++ Python GoogleCNTK [73] C++ MicrosoftKeras [74] Theano Tensor Flow MITdl4j [75] Java Skymind Engineering
DeeBNET [76 77] MATLAB Information Technology DepartmentAmirkabir University of Technology
has been achieved in the image classification and analysisfield As an advanced engineering of the AlexNet the papertitled ldquoGoing Deeper with Convolutionsrdquo by Szegedy [115]introduced the GoogleNet model This model contains amuch deeper network than AlexNet Sequentially ResNet[116] Inception [117] Inception-v4 Inception-ResNet [118]and a few other models have recently been introduced
Later directly or with some advanced modificationthese DNN models have been adapted for biomedical imageanalysis In 2015 Fonseca et al [81] classified breast densityusing CNN techniques CNN requires a sufficient amountof data to train the system It is always very difficult tofind a sufficient amount of medical data for training a CNNmodel A pretrained CNN model with some fine tuning canbe used rather than create a model from scratch [119] Theauthors of [119] did not perform their experiments on a breastcancer image dataset however they have performed theirexperiments on three different medical datasets with layer-wise training and claimed that ldquoretrained CNN along withadequate training can provide better or at least the sameamount of performancerdquo
The Deep Belief Network (DBN) is another branch of theDeep Neural Network which mainly consists of RestrictedBoltzmann Machine (RBM) techniques The DBN methodwas first utilized for supervised image classification by Liu etal [120] After that Abdel-Zaher and Eldeib utilized the DBNmethod for breast image classification [121] This field is stillnot fully explored for breast image classification yet Zhanget al utilized both RBM and Point-Wise Gated RBM (PRBM)for shear-wave electrography image classification where thedataset contains 227 images [97]Their achieved classificationAccuracy Sensitivity and Specificity are 9340 8860 and9710 respectively Tables 9 10 and 11 have summarized themost recent work for breast image classification along withsome pioneer work on CNN
313 Logic Based Algorithm A Logic Based algorithm isa very popular and effective classification method whichfollows the tree structure principle and logical argument asshown in Figure 16 This algorithm classifies instances based
on the featurersquos values Along with other criteria a decision-tree based algorithm contains the following features
(i) Root node a root node contains no incoming nodeand it may or may not contain any outgoing edge
(ii) Splitting splitting is the process of subdividing a set ofcases into a particular group Normally the followingcriteria are maintained for the splitting
(a) information gain(b) Gini index(c) chi squared
(iii) Decision node(iv) Leafterminal node this kind of node has exactly one
incoming edge and no outgoing edgeThe tree alwaysterminates here with a decision
(v) Pruning pruning is a process of removing subtreesfrom the tree Pruning performs to reduce the over-fitting problem Two kinds of pruning techniques areavailable
(a) prepruning(b) postpruning
Among all the tree based algorithms IterativeDichotomiser 3 (ID3) can be considered as a pioneerproposed by Quinlan [149] The problem of the ID3algorithm is to find the optimal solution which is very muchprone towards overfitting To overcome the limitation of theID3 algorithm the C45 algorithm has been introduced byQuinlan [150] where a pruning method has been introducedto control the overfitting problem Pritom et al [151] classifiedthe Wisconsin breast dataset where they utilized 35 featuresThey have obtained 7630 Accuracy 7510 False PositiveRate and ROC score 0745 when they ranked the featuresWithout ranking the features they obtained 7370Accuracy5070 False Positive Rate and ROC score value 5280 Asriet al [152] utilized the C45 algorithm for the Wisconsin
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
(a) Self-training(b) Graph Based(c) S3V3(d) Multiview(e) Generative model
Among the different classification performance proper-ties this matrix will provide following parameters
(i) Recall is defined as Recall = TP(TP + FN)(ii) Precision is defined as Precision = TP(TP + FP)(iii) Specificity is defined as Specificity = TN(TN + FP)(iv) Accuracy is defined as ACC = (TP+TN)(TP+TN+
FP + FN)(v) F-1 score is defined as 1198651 = (2 times Recall)(2 times Recall +
FP + FN)(vi) Matthew Correlation Coefficient (MCC) MCC is a
performance parameter of a binary classifier in therange minus1 to +1 If the MCC values trend moretowards +1 the classifier gives a more accurate classi-fier and the opposite condition will occur if the valueof theMCC trend towards theminus1MCCcanbe definedas
MCC
= TP times TN minus FP times FNradic(TP + FP) (TP + FN) (TN + FP) (TN + FP) (1)
3 Performance of Different Classifier Modelon Breast Images Dataset
Based on Supervised Semisupervised and Unsupervisedmethods different research groups have been performedclassification operation on different image database In thissection we have summarized few of the works of breast imageclassification
31 Performance Based on Supervised Learning In super-vised learning a general hypothesis is established based onexternally supplied instances to produce future predictionFor the supervised classification task features are extractedor automatically crafted from the available dataset and eachsample is mapped to a dedicated class With the help of thefeatures and their levels a hypothesis is created Based on thehypothesis unknown data are classified [55]
Figure 10 represents an overall supervised classifier archi-tecture In general the whole dataset is split into trainingand testing parts To validate the data some time dataare also split into a validation part as well After the datasplitting themost important part is to find out the appropriatefeatures to classify the data with the utmost AccuracyFinding the features can be classified into two categorieslocally and globally crafted Locally crafted means that thismethod requires a hand-held exercise to find out the featureswhereas globally craftedmeans that a kernelmethod has beenintroduced for the feature extraction Handcrafted featurescan be prioritized whereas Global Feature selection does nothave this luxury
311 Conventional Neural Network The Neural Network(NN) concept comes from the working principle of thehuman brain A biological neuron consists of the followingfour parts
8 Computational and Mathematical Methods in Medicine
Classifier model
Imagedatabase
Traintestdata splitting Locally
craftedGloballycrafted
Hand crafting
Kernel basedcrafting
Featureprioritization
Conventionalclassifier
DNNclassifier
Evaluationmatrix
Classifieddata
Feature collection
Ensemble learning
Figure 10 A generalized supervised classifier model
Nucleus
Axon
Cell body
Dendrites
Figure 11 A model of a biological neuron
Dendrites collect signals and axons carry the signal to thenext dendrite after processing by the cell body as shown inFigure 11 Using the neuronworking principle the perceptronmodel was proposed by Rosenblatt in 1957 [56] A single-layer perceptron linearly combines the input signal and givesa decision based on a threshold function Based on theworking principle and with some advanced mechanism andengineering NNmethods have established a strong footprintin many problem-solving issues Figure 12 shows the basicworking principle of NN techniques
In the NN model the input data X = 1199090 1199091 119909119873 isfirst multiplied by the weight dataW = 1199080 1199081 119908119873 andthen the output is calculated using
Y = g (sum) wheresum = W sdot X (2)
Function g is known as the activation function Thisfunction can be any threshold value or Sigmoid or hyperbolicand so forth In the early stages feed-forwardNeuralNetworktechniques were introduced [57] lately the backpropagationmethod has been invented to utilize the error information toimprove the system performance [58 59]
The history of breast image classification by NN is a longone To the best of my knowledge a lot of the pioneer work
yg
x0
x1
xNminus1
xN
w0
w1
wNminus1
wN
Figure 12Working principle of a simpleNeuralNetwork technique
was performed by Dawson et al in 1991 [60] Since then NNhas been utilized as one of the strong tools for breast imageclassification We have summarized some of the work relatedto NN and breast image classification in Tables 5 6 and 7
312 Deep Neural Network Deep Neural Network (DNN) isa state-of-the-art concept where conventional NN techniqueshave been utilized with advanced engineering It is foundthat conventional NNs have difficulties in solving complexproblems whereas DNNs solve them with utmost PrecisionHowever DNNs suffer from more time and computationalcomplexity than the conventional NN
Convolutional Neural Network A CNN model is the combi-nation of a few intermediate mathematical structures Thisintermediatemathematical structure creates or helps to createdifferent layers
(i) Convolutional Layer Among all the other layers theconvolutional layer is considered as the most important partfor a CNN model and can be considered as the backbone of
Computational and Mathematical Methods in Medicine 9
Table 5 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Rajakeerthana et al [42] (1) GLCM GLDM SRDMNGLCM GLRM Mammogram 322 (1)The classifier achieved 9920
Accuracy
Lessa and Marengoni [43](1)Mean Median StandardDeviation Skewness KurtosisEntropy Range
Wan et al [44] (1) ALBP (2) BBLBP OCM 46(1) Achieved Sensitivity and Specificityare 100 and 8520 respectively(2) ROC value obtained 0959
Chen et al [40] (1) 19 BI-RADS features havebeen used Ultrasound 238
(1) Chi squared method has beenutilized for the feature selection(2) Achieved Accuracy Sensitivity andSpecificity are 9610 9670 and9570 respectively
de Lima et al [45] (1) Total 416 features have beenused Mammogram 355
(1)Multiresolution wavelet and Zernikemoment have been utilized for thefeature extraction
Abirami et al [46](1) 12 statistical measures such asMean Median and Max havebeen utilized as the features
Mammogram 322
(1)Wavelet transform has been utilizedfor the feature extraction(2)The achieved Accuracy Sensitivityand Specificity are 9550 9500 and9600 respectively
El Atlas et al [47] (1) 13 morphological featureshave been utilized Mammogram 410
(1) Firstly the edge information hasbeen utilized for the mass segmentationand then the morphological featureswere extracted(2) Achieved best Accuracy is 975
Table 6 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Alharbi et al [48] (1) 49 features havebeen utilized Mammogram 1100
(1) Five feature selection methods Fisher scoreMinimum Redundancy-Maximum Relevance Relief-fSequential Forward Feature Selection and GeneticAlgorithm have been used(2) Achieved Accuracy Sensitivity and specificity are9420 9836 and 9927 respectively
Peng et al [49](1)Haralick andTamura features havebeen utilized
Mammogram 322
(1) Feature reduction has been performed byRough-Set theory and selected 5 prioritized features(2)The best Accuracy Sensitivity and Specificityachieved were 9600 9860 and 8930
Jalalian et al [50] (1) GLCM Mammogram(1)The obtained classifier Accuracy Sensitivity andSpecificity are 9520 9240 and 9800respectively(2) Compactness
Li et al [51](1) Four featurevectors have beencalculated
Mammogram 322
(1) 2D contour of breast mass in mammography hasbeen converted into 1D signature(2) NN techniques achieved Accuracy is 9960 whenRMS slope is utilized
Chen et al [52] (1) Autocorrelationfeatures Ultrasound 242 (1)The overall achieved Accuracy Sensitivity and
Specificity are 9500 9800 and 93 respectively
Chen et al [53] (1) Autocorrelationfeatures Ultrasound 1020 (1)The obtained ROC area is 09840 plusmn 00072
10 Computational and Mathematical Methods in Medicine
Table 7 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Chen et al [61]
(1) Variance Contrast of WaveletCoefficient Ultrasound 242 (1)The achieved ROC curve 09396 plusmn 00183(2) Autocorrelation of WaveletCoefficient
Silva et al [62](1) 22 different morphologicalfeatures such as convexity andlobulation have been utilized
Ultrasound mdash (1)The best obtained Accuracy and ROCcurve are 9698 and 098 respectively
Saritas [63](1) Age of patient (2)massshape (3)mass border (4)Massdensity (5) BIRADS Mammogram mdash
(1) Disease prediction rate is 905(2) Neural Network utilized 5 neurons ininput layers and one hidden layer
Lopez-Melendez etal [64]
(1) Area perimeter etc havebeen utilized Mammogram 322 (1)The achieved Sensitivity and Specificity
are 9629 and 9900 respectively
themodel A kernel of size119898times119899 is scanned through the inputdata for the convolutional operation which ensures the localconnectivity and weight sharing property
(ii) Stride and Padding In the convolutional operation afilter scans through the input matrices In each step howmuch position a kernel filter moves through the matrixis known as the stride By default stride keeps to 1 Withinappropriate selection of the stride the model can lose theborder information To overcome this issue themodel utilizesextra rows and columns at the end of the matrices and theseadded rows and columns contain all 0s This adding of extrarows and columns which contain only zero value is known aszero padding
(iii) Nonlinear Operation The output of each of the kerneloperations is passed through a rectifier function such as Rec-tified Linear Unit (ReLU) Leaky-ReLU TanH and SigmoidThe Sigmoid function can be defined as
120590 (119909) = 1(1 + expminus119909) (3)
and the tanh function can be defined as
tanh (119909) = (exp119909 minus expminus119909)(exp119909 + expminus119909) (4)
However the most effective rectifier is ReLU The ReLUmethod converts all the information into zero if it is less thanor equal to zero and passes all the other data as is shown inFigure 13
120590 (119909) = max (0 119909) (5)
Another important nonlinear function is Leaky-RelU
where 120572 is predetermined parameter which can be varied togive a better model
minus3 minus2 minus1 0 1 2 3
1
2
3
InputO
utpu
t
Figure 13 ReLU Operation
(iv) Subsampling Subsampling is the procedure of reducingthe dimensionality of each of the feature maps of a particularlayer this operation is also known as a pooling operationActually it reduces the amount of feature information fromthe overall data By doing so it reduces the overall computa-tional complexity of themodel To do this 119904times119904 patch units areutilized The two most popular pooling methods are
(a) Max-Pooling
(b) Average Pooling
In Max-Pooling only the maximum values within a partic-ular kernel size are selected for further calculation Consideran example of a 16 times 16 image as shown in Figure 14 A 2 by2 kernel is applied to the whole image 4 blocks in total andproduces a 4 times 4 output image For each block of four valueswe have selected the maximum For instance from blocksone two three and four maximum values 4 40 13 and 8are selected respectively as they are the maximum in thatblock For the Average Pooling operation each kernel givesthe output as average
(v) Dropout Regularization of the weight can reduce theoutfitting problem Randomly removing some neurons can
Computational and Mathematical Methods in Medicine 11
Figure 15 Work-flow of a Convolutional Neural Network
regularize the overfilling problem The technique of ran-domly removing neurons from the network is known asdropout
(vi) Soft-Max Layer This layer contains normalized expo-nential functions to calculate the loss function for the dataclassification
Figure 15 shows a generalized CNN model for the imageclassificationAll the neurons of themost immediate layer of afully connected layer are completely connected with the fullyconnected layer like a conventional Neural Network Let119891119897minus1119895represent the 119895th feature map at the layer 119897minus1The 119895th featuremap at the layer 119897 can be represented as
where119873119897minus119897 represents the number of featuremaps at the 119897minus1thlayer 119896119894119895 represents the kernel function and 119887119897119895 represents thebias at 119897 where 120590 performs a nonlinear function operationThe layer before the Soft-Max Layer can be represented as
Let 119901 = 1 represent Benign class and 119901 = 2 represent theMalignant class The cross-entropy loss of the above functioncan be calculated as
119871119901 = minus ln (119910119901) (10)
Whichever group experiences a large loss value themodel will consider the other group as predicted class
A difficult part of working on DNN is that it requiresa specialized software package for the data analysis Fewresearch groups have been working on how effectively datacan be analyzed by DNN from different perspectives and thedemand Table 8 summarizes some of the software which isavailable for DNN analysis
The history of the CNN and its use for biomedical imageanalysis is a long one Fukushima first introduced a CNNnamed ldquonecognitronrdquo which has the ability to recognizestimulus patterns with a few shifting variances [113] Tothe best of our knowledge Wu et al first classified a setof mammogram images into malignant and benign classesusing a CNN model [78] In their proposed model they onlyutilized one hidden layer After that in 1996 Sahiner et alutilized CNNmodel to classify mass and normal breast tissueand achieved ROC scores of 087 [79] In 2002 Lo et alutilized aMultiple Circular Path CNN (MCPCNN) for tumoridentification from mammogram images and obtained ROCscores of around 089 After an absence of investigation ofthe CNN model this model regained its momentum afterthe work of Krizhevsky et al [114] Their proposed model isknown as AlexNet After this work a revolutionary change
12 Computational and Mathematical Methods in Medicine
Table 8 Available software for deep learning analysis
Software Interface and backend Provider
Caffe [65 66] Python MATLAB C++ Berkeley Vision and Learning CentreUniversity of California Berkeley
Torch [67] C LuaJIT
MatConvNet [68 69] MATLAB C Visual Geometry Group Department ofEngineering University of Oxford
Theano [70 71] Python Montreal Institute for Learning AlgorithmsUniversity of Montreal
TensorFlows [72] C++ Python GoogleCNTK [73] C++ MicrosoftKeras [74] Theano Tensor Flow MITdl4j [75] Java Skymind Engineering
DeeBNET [76 77] MATLAB Information Technology DepartmentAmirkabir University of Technology
has been achieved in the image classification and analysisfield As an advanced engineering of the AlexNet the papertitled ldquoGoing Deeper with Convolutionsrdquo by Szegedy [115]introduced the GoogleNet model This model contains amuch deeper network than AlexNet Sequentially ResNet[116] Inception [117] Inception-v4 Inception-ResNet [118]and a few other models have recently been introduced
Later directly or with some advanced modificationthese DNN models have been adapted for biomedical imageanalysis In 2015 Fonseca et al [81] classified breast densityusing CNN techniques CNN requires a sufficient amountof data to train the system It is always very difficult tofind a sufficient amount of medical data for training a CNNmodel A pretrained CNN model with some fine tuning canbe used rather than create a model from scratch [119] Theauthors of [119] did not perform their experiments on a breastcancer image dataset however they have performed theirexperiments on three different medical datasets with layer-wise training and claimed that ldquoretrained CNN along withadequate training can provide better or at least the sameamount of performancerdquo
The Deep Belief Network (DBN) is another branch of theDeep Neural Network which mainly consists of RestrictedBoltzmann Machine (RBM) techniques The DBN methodwas first utilized for supervised image classification by Liu etal [120] After that Abdel-Zaher and Eldeib utilized the DBNmethod for breast image classification [121] This field is stillnot fully explored for breast image classification yet Zhanget al utilized both RBM and Point-Wise Gated RBM (PRBM)for shear-wave electrography image classification where thedataset contains 227 images [97]Their achieved classificationAccuracy Sensitivity and Specificity are 9340 8860 and9710 respectively Tables 9 10 and 11 have summarized themost recent work for breast image classification along withsome pioneer work on CNN
313 Logic Based Algorithm A Logic Based algorithm isa very popular and effective classification method whichfollows the tree structure principle and logical argument asshown in Figure 16 This algorithm classifies instances based
on the featurersquos values Along with other criteria a decision-tree based algorithm contains the following features
(i) Root node a root node contains no incoming nodeand it may or may not contain any outgoing edge
(ii) Splitting splitting is the process of subdividing a set ofcases into a particular group Normally the followingcriteria are maintained for the splitting
(a) information gain(b) Gini index(c) chi squared
(iii) Decision node(iv) Leafterminal node this kind of node has exactly one
incoming edge and no outgoing edgeThe tree alwaysterminates here with a decision
(v) Pruning pruning is a process of removing subtreesfrom the tree Pruning performs to reduce the over-fitting problem Two kinds of pruning techniques areavailable
(a) prepruning(b) postpruning
Among all the tree based algorithms IterativeDichotomiser 3 (ID3) can be considered as a pioneerproposed by Quinlan [149] The problem of the ID3algorithm is to find the optimal solution which is very muchprone towards overfitting To overcome the limitation of theID3 algorithm the C45 algorithm has been introduced byQuinlan [150] where a pruning method has been introducedto control the overfitting problem Pritom et al [151] classifiedthe Wisconsin breast dataset where they utilized 35 featuresThey have obtained 7630 Accuracy 7510 False PositiveRate and ROC score 0745 when they ranked the featuresWithout ranking the features they obtained 7370Accuracy5070 False Positive Rate and ROC score value 5280 Asriet al [152] utilized the C45 algorithm for the Wisconsin
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
8 Computational and Mathematical Methods in Medicine
Classifier model
Imagedatabase
Traintestdata splitting Locally
craftedGloballycrafted
Hand crafting
Kernel basedcrafting
Featureprioritization
Conventionalclassifier
DNNclassifier
Evaluationmatrix
Classifieddata
Feature collection
Ensemble learning
Figure 10 A generalized supervised classifier model
Nucleus
Axon
Cell body
Dendrites
Figure 11 A model of a biological neuron
Dendrites collect signals and axons carry the signal to thenext dendrite after processing by the cell body as shown inFigure 11 Using the neuronworking principle the perceptronmodel was proposed by Rosenblatt in 1957 [56] A single-layer perceptron linearly combines the input signal and givesa decision based on a threshold function Based on theworking principle and with some advanced mechanism andengineering NNmethods have established a strong footprintin many problem-solving issues Figure 12 shows the basicworking principle of NN techniques
In the NN model the input data X = 1199090 1199091 119909119873 isfirst multiplied by the weight dataW = 1199080 1199081 119908119873 andthen the output is calculated using
Y = g (sum) wheresum = W sdot X (2)
Function g is known as the activation function Thisfunction can be any threshold value or Sigmoid or hyperbolicand so forth In the early stages feed-forwardNeuralNetworktechniques were introduced [57] lately the backpropagationmethod has been invented to utilize the error information toimprove the system performance [58 59]
The history of breast image classification by NN is a longone To the best of my knowledge a lot of the pioneer work
yg
x0
x1
xNminus1
xN
w0
w1
wNminus1
wN
Figure 12Working principle of a simpleNeuralNetwork technique
was performed by Dawson et al in 1991 [60] Since then NNhas been utilized as one of the strong tools for breast imageclassification We have summarized some of the work relatedto NN and breast image classification in Tables 5 6 and 7
312 Deep Neural Network Deep Neural Network (DNN) isa state-of-the-art concept where conventional NN techniqueshave been utilized with advanced engineering It is foundthat conventional NNs have difficulties in solving complexproblems whereas DNNs solve them with utmost PrecisionHowever DNNs suffer from more time and computationalcomplexity than the conventional NN
Convolutional Neural Network A CNN model is the combi-nation of a few intermediate mathematical structures Thisintermediatemathematical structure creates or helps to createdifferent layers
(i) Convolutional Layer Among all the other layers theconvolutional layer is considered as the most important partfor a CNN model and can be considered as the backbone of
Computational and Mathematical Methods in Medicine 9
Table 5 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Rajakeerthana et al [42] (1) GLCM GLDM SRDMNGLCM GLRM Mammogram 322 (1)The classifier achieved 9920
Accuracy
Lessa and Marengoni [43](1)Mean Median StandardDeviation Skewness KurtosisEntropy Range
Wan et al [44] (1) ALBP (2) BBLBP OCM 46(1) Achieved Sensitivity and Specificityare 100 and 8520 respectively(2) ROC value obtained 0959
Chen et al [40] (1) 19 BI-RADS features havebeen used Ultrasound 238
(1) Chi squared method has beenutilized for the feature selection(2) Achieved Accuracy Sensitivity andSpecificity are 9610 9670 and9570 respectively
de Lima et al [45] (1) Total 416 features have beenused Mammogram 355
(1)Multiresolution wavelet and Zernikemoment have been utilized for thefeature extraction
Abirami et al [46](1) 12 statistical measures such asMean Median and Max havebeen utilized as the features
Mammogram 322
(1)Wavelet transform has been utilizedfor the feature extraction(2)The achieved Accuracy Sensitivityand Specificity are 9550 9500 and9600 respectively
El Atlas et al [47] (1) 13 morphological featureshave been utilized Mammogram 410
(1) Firstly the edge information hasbeen utilized for the mass segmentationand then the morphological featureswere extracted(2) Achieved best Accuracy is 975
Table 6 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Alharbi et al [48] (1) 49 features havebeen utilized Mammogram 1100
(1) Five feature selection methods Fisher scoreMinimum Redundancy-Maximum Relevance Relief-fSequential Forward Feature Selection and GeneticAlgorithm have been used(2) Achieved Accuracy Sensitivity and specificity are9420 9836 and 9927 respectively
Peng et al [49](1)Haralick andTamura features havebeen utilized
Mammogram 322
(1) Feature reduction has been performed byRough-Set theory and selected 5 prioritized features(2)The best Accuracy Sensitivity and Specificityachieved were 9600 9860 and 8930
Jalalian et al [50] (1) GLCM Mammogram(1)The obtained classifier Accuracy Sensitivity andSpecificity are 9520 9240 and 9800respectively(2) Compactness
Li et al [51](1) Four featurevectors have beencalculated
Mammogram 322
(1) 2D contour of breast mass in mammography hasbeen converted into 1D signature(2) NN techniques achieved Accuracy is 9960 whenRMS slope is utilized
Chen et al [52] (1) Autocorrelationfeatures Ultrasound 242 (1)The overall achieved Accuracy Sensitivity and
Specificity are 9500 9800 and 93 respectively
Chen et al [53] (1) Autocorrelationfeatures Ultrasound 1020 (1)The obtained ROC area is 09840 plusmn 00072
10 Computational and Mathematical Methods in Medicine
Table 7 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Chen et al [61]
(1) Variance Contrast of WaveletCoefficient Ultrasound 242 (1)The achieved ROC curve 09396 plusmn 00183(2) Autocorrelation of WaveletCoefficient
Silva et al [62](1) 22 different morphologicalfeatures such as convexity andlobulation have been utilized
Ultrasound mdash (1)The best obtained Accuracy and ROCcurve are 9698 and 098 respectively
Saritas [63](1) Age of patient (2)massshape (3)mass border (4)Massdensity (5) BIRADS Mammogram mdash
(1) Disease prediction rate is 905(2) Neural Network utilized 5 neurons ininput layers and one hidden layer
Lopez-Melendez etal [64]
(1) Area perimeter etc havebeen utilized Mammogram 322 (1)The achieved Sensitivity and Specificity
are 9629 and 9900 respectively
themodel A kernel of size119898times119899 is scanned through the inputdata for the convolutional operation which ensures the localconnectivity and weight sharing property
(ii) Stride and Padding In the convolutional operation afilter scans through the input matrices In each step howmuch position a kernel filter moves through the matrixis known as the stride By default stride keeps to 1 Withinappropriate selection of the stride the model can lose theborder information To overcome this issue themodel utilizesextra rows and columns at the end of the matrices and theseadded rows and columns contain all 0s This adding of extrarows and columns which contain only zero value is known aszero padding
(iii) Nonlinear Operation The output of each of the kerneloperations is passed through a rectifier function such as Rec-tified Linear Unit (ReLU) Leaky-ReLU TanH and SigmoidThe Sigmoid function can be defined as
120590 (119909) = 1(1 + expminus119909) (3)
and the tanh function can be defined as
tanh (119909) = (exp119909 minus expminus119909)(exp119909 + expminus119909) (4)
However the most effective rectifier is ReLU The ReLUmethod converts all the information into zero if it is less thanor equal to zero and passes all the other data as is shown inFigure 13
120590 (119909) = max (0 119909) (5)
Another important nonlinear function is Leaky-RelU
where 120572 is predetermined parameter which can be varied togive a better model
minus3 minus2 minus1 0 1 2 3
1
2
3
InputO
utpu
t
Figure 13 ReLU Operation
(iv) Subsampling Subsampling is the procedure of reducingthe dimensionality of each of the feature maps of a particularlayer this operation is also known as a pooling operationActually it reduces the amount of feature information fromthe overall data By doing so it reduces the overall computa-tional complexity of themodel To do this 119904times119904 patch units areutilized The two most popular pooling methods are
(a) Max-Pooling
(b) Average Pooling
In Max-Pooling only the maximum values within a partic-ular kernel size are selected for further calculation Consideran example of a 16 times 16 image as shown in Figure 14 A 2 by2 kernel is applied to the whole image 4 blocks in total andproduces a 4 times 4 output image For each block of four valueswe have selected the maximum For instance from blocksone two three and four maximum values 4 40 13 and 8are selected respectively as they are the maximum in thatblock For the Average Pooling operation each kernel givesthe output as average
(v) Dropout Regularization of the weight can reduce theoutfitting problem Randomly removing some neurons can
Computational and Mathematical Methods in Medicine 11
Figure 15 Work-flow of a Convolutional Neural Network
regularize the overfilling problem The technique of ran-domly removing neurons from the network is known asdropout
(vi) Soft-Max Layer This layer contains normalized expo-nential functions to calculate the loss function for the dataclassification
Figure 15 shows a generalized CNN model for the imageclassificationAll the neurons of themost immediate layer of afully connected layer are completely connected with the fullyconnected layer like a conventional Neural Network Let119891119897minus1119895represent the 119895th feature map at the layer 119897minus1The 119895th featuremap at the layer 119897 can be represented as
where119873119897minus119897 represents the number of featuremaps at the 119897minus1thlayer 119896119894119895 represents the kernel function and 119887119897119895 represents thebias at 119897 where 120590 performs a nonlinear function operationThe layer before the Soft-Max Layer can be represented as
Let 119901 = 1 represent Benign class and 119901 = 2 represent theMalignant class The cross-entropy loss of the above functioncan be calculated as
119871119901 = minus ln (119910119901) (10)
Whichever group experiences a large loss value themodel will consider the other group as predicted class
A difficult part of working on DNN is that it requiresa specialized software package for the data analysis Fewresearch groups have been working on how effectively datacan be analyzed by DNN from different perspectives and thedemand Table 8 summarizes some of the software which isavailable for DNN analysis
The history of the CNN and its use for biomedical imageanalysis is a long one Fukushima first introduced a CNNnamed ldquonecognitronrdquo which has the ability to recognizestimulus patterns with a few shifting variances [113] Tothe best of our knowledge Wu et al first classified a setof mammogram images into malignant and benign classesusing a CNN model [78] In their proposed model they onlyutilized one hidden layer After that in 1996 Sahiner et alutilized CNNmodel to classify mass and normal breast tissueand achieved ROC scores of 087 [79] In 2002 Lo et alutilized aMultiple Circular Path CNN (MCPCNN) for tumoridentification from mammogram images and obtained ROCscores of around 089 After an absence of investigation ofthe CNN model this model regained its momentum afterthe work of Krizhevsky et al [114] Their proposed model isknown as AlexNet After this work a revolutionary change
12 Computational and Mathematical Methods in Medicine
Table 8 Available software for deep learning analysis
Software Interface and backend Provider
Caffe [65 66] Python MATLAB C++ Berkeley Vision and Learning CentreUniversity of California Berkeley
Torch [67] C LuaJIT
MatConvNet [68 69] MATLAB C Visual Geometry Group Department ofEngineering University of Oxford
Theano [70 71] Python Montreal Institute for Learning AlgorithmsUniversity of Montreal
TensorFlows [72] C++ Python GoogleCNTK [73] C++ MicrosoftKeras [74] Theano Tensor Flow MITdl4j [75] Java Skymind Engineering
DeeBNET [76 77] MATLAB Information Technology DepartmentAmirkabir University of Technology
has been achieved in the image classification and analysisfield As an advanced engineering of the AlexNet the papertitled ldquoGoing Deeper with Convolutionsrdquo by Szegedy [115]introduced the GoogleNet model This model contains amuch deeper network than AlexNet Sequentially ResNet[116] Inception [117] Inception-v4 Inception-ResNet [118]and a few other models have recently been introduced
Later directly or with some advanced modificationthese DNN models have been adapted for biomedical imageanalysis In 2015 Fonseca et al [81] classified breast densityusing CNN techniques CNN requires a sufficient amountof data to train the system It is always very difficult tofind a sufficient amount of medical data for training a CNNmodel A pretrained CNN model with some fine tuning canbe used rather than create a model from scratch [119] Theauthors of [119] did not perform their experiments on a breastcancer image dataset however they have performed theirexperiments on three different medical datasets with layer-wise training and claimed that ldquoretrained CNN along withadequate training can provide better or at least the sameamount of performancerdquo
The Deep Belief Network (DBN) is another branch of theDeep Neural Network which mainly consists of RestrictedBoltzmann Machine (RBM) techniques The DBN methodwas first utilized for supervised image classification by Liu etal [120] After that Abdel-Zaher and Eldeib utilized the DBNmethod for breast image classification [121] This field is stillnot fully explored for breast image classification yet Zhanget al utilized both RBM and Point-Wise Gated RBM (PRBM)for shear-wave electrography image classification where thedataset contains 227 images [97]Their achieved classificationAccuracy Sensitivity and Specificity are 9340 8860 and9710 respectively Tables 9 10 and 11 have summarized themost recent work for breast image classification along withsome pioneer work on CNN
313 Logic Based Algorithm A Logic Based algorithm isa very popular and effective classification method whichfollows the tree structure principle and logical argument asshown in Figure 16 This algorithm classifies instances based
on the featurersquos values Along with other criteria a decision-tree based algorithm contains the following features
(i) Root node a root node contains no incoming nodeand it may or may not contain any outgoing edge
(ii) Splitting splitting is the process of subdividing a set ofcases into a particular group Normally the followingcriteria are maintained for the splitting
(a) information gain(b) Gini index(c) chi squared
(iii) Decision node(iv) Leafterminal node this kind of node has exactly one
incoming edge and no outgoing edgeThe tree alwaysterminates here with a decision
(v) Pruning pruning is a process of removing subtreesfrom the tree Pruning performs to reduce the over-fitting problem Two kinds of pruning techniques areavailable
(a) prepruning(b) postpruning
Among all the tree based algorithms IterativeDichotomiser 3 (ID3) can be considered as a pioneerproposed by Quinlan [149] The problem of the ID3algorithm is to find the optimal solution which is very muchprone towards overfitting To overcome the limitation of theID3 algorithm the C45 algorithm has been introduced byQuinlan [150] where a pruning method has been introducedto control the overfitting problem Pritom et al [151] classifiedthe Wisconsin breast dataset where they utilized 35 featuresThey have obtained 7630 Accuracy 7510 False PositiveRate and ROC score 0745 when they ranked the featuresWithout ranking the features they obtained 7370Accuracy5070 False Positive Rate and ROC score value 5280 Asriet al [152] utilized the C45 algorithm for the Wisconsin
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Wan et al [44] (1) ALBP (2) BBLBP OCM 46(1) Achieved Sensitivity and Specificityare 100 and 8520 respectively(2) ROC value obtained 0959
Chen et al [40] (1) 19 BI-RADS features havebeen used Ultrasound 238
(1) Chi squared method has beenutilized for the feature selection(2) Achieved Accuracy Sensitivity andSpecificity are 9610 9670 and9570 respectively
de Lima et al [45] (1) Total 416 features have beenused Mammogram 355
(1)Multiresolution wavelet and Zernikemoment have been utilized for thefeature extraction
Abirami et al [46](1) 12 statistical measures such asMean Median and Max havebeen utilized as the features
Mammogram 322
(1)Wavelet transform has been utilizedfor the feature extraction(2)The achieved Accuracy Sensitivityand Specificity are 9550 9500 and9600 respectively
El Atlas et al [47] (1) 13 morphological featureshave been utilized Mammogram 410
(1) Firstly the edge information hasbeen utilized for the mass segmentationand then the morphological featureswere extracted(2) Achieved best Accuracy is 975
Table 6 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Alharbi et al [48] (1) 49 features havebeen utilized Mammogram 1100
(1) Five feature selection methods Fisher scoreMinimum Redundancy-Maximum Relevance Relief-fSequential Forward Feature Selection and GeneticAlgorithm have been used(2) Achieved Accuracy Sensitivity and specificity are9420 9836 and 9927 respectively
Peng et al [49](1)Haralick andTamura features havebeen utilized
Mammogram 322
(1) Feature reduction has been performed byRough-Set theory and selected 5 prioritized features(2)The best Accuracy Sensitivity and Specificityachieved were 9600 9860 and 8930
Jalalian et al [50] (1) GLCM Mammogram(1)The obtained classifier Accuracy Sensitivity andSpecificity are 9520 9240 and 9800respectively(2) Compactness
Li et al [51](1) Four featurevectors have beencalculated
Mammogram 322
(1) 2D contour of breast mass in mammography hasbeen converted into 1D signature(2) NN techniques achieved Accuracy is 9960 whenRMS slope is utilized
Chen et al [52] (1) Autocorrelationfeatures Ultrasound 242 (1)The overall achieved Accuracy Sensitivity and
Specificity are 9500 9800 and 93 respectively
Chen et al [53] (1) Autocorrelationfeatures Ultrasound 1020 (1)The obtained ROC area is 09840 plusmn 00072
10 Computational and Mathematical Methods in Medicine
Table 7 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Chen et al [61]
(1) Variance Contrast of WaveletCoefficient Ultrasound 242 (1)The achieved ROC curve 09396 plusmn 00183(2) Autocorrelation of WaveletCoefficient
Silva et al [62](1) 22 different morphologicalfeatures such as convexity andlobulation have been utilized
Ultrasound mdash (1)The best obtained Accuracy and ROCcurve are 9698 and 098 respectively
Saritas [63](1) Age of patient (2)massshape (3)mass border (4)Massdensity (5) BIRADS Mammogram mdash
(1) Disease prediction rate is 905(2) Neural Network utilized 5 neurons ininput layers and one hidden layer
Lopez-Melendez etal [64]
(1) Area perimeter etc havebeen utilized Mammogram 322 (1)The achieved Sensitivity and Specificity
are 9629 and 9900 respectively
themodel A kernel of size119898times119899 is scanned through the inputdata for the convolutional operation which ensures the localconnectivity and weight sharing property
(ii) Stride and Padding In the convolutional operation afilter scans through the input matrices In each step howmuch position a kernel filter moves through the matrixis known as the stride By default stride keeps to 1 Withinappropriate selection of the stride the model can lose theborder information To overcome this issue themodel utilizesextra rows and columns at the end of the matrices and theseadded rows and columns contain all 0s This adding of extrarows and columns which contain only zero value is known aszero padding
(iii) Nonlinear Operation The output of each of the kerneloperations is passed through a rectifier function such as Rec-tified Linear Unit (ReLU) Leaky-ReLU TanH and SigmoidThe Sigmoid function can be defined as
120590 (119909) = 1(1 + expminus119909) (3)
and the tanh function can be defined as
tanh (119909) = (exp119909 minus expminus119909)(exp119909 + expminus119909) (4)
However the most effective rectifier is ReLU The ReLUmethod converts all the information into zero if it is less thanor equal to zero and passes all the other data as is shown inFigure 13
120590 (119909) = max (0 119909) (5)
Another important nonlinear function is Leaky-RelU
where 120572 is predetermined parameter which can be varied togive a better model
minus3 minus2 minus1 0 1 2 3
1
2
3
InputO
utpu
t
Figure 13 ReLU Operation
(iv) Subsampling Subsampling is the procedure of reducingthe dimensionality of each of the feature maps of a particularlayer this operation is also known as a pooling operationActually it reduces the amount of feature information fromthe overall data By doing so it reduces the overall computa-tional complexity of themodel To do this 119904times119904 patch units areutilized The two most popular pooling methods are
(a) Max-Pooling
(b) Average Pooling
In Max-Pooling only the maximum values within a partic-ular kernel size are selected for further calculation Consideran example of a 16 times 16 image as shown in Figure 14 A 2 by2 kernel is applied to the whole image 4 blocks in total andproduces a 4 times 4 output image For each block of four valueswe have selected the maximum For instance from blocksone two three and four maximum values 4 40 13 and 8are selected respectively as they are the maximum in thatblock For the Average Pooling operation each kernel givesthe output as average
(v) Dropout Regularization of the weight can reduce theoutfitting problem Randomly removing some neurons can
Computational and Mathematical Methods in Medicine 11
Figure 15 Work-flow of a Convolutional Neural Network
regularize the overfilling problem The technique of ran-domly removing neurons from the network is known asdropout
(vi) Soft-Max Layer This layer contains normalized expo-nential functions to calculate the loss function for the dataclassification
Figure 15 shows a generalized CNN model for the imageclassificationAll the neurons of themost immediate layer of afully connected layer are completely connected with the fullyconnected layer like a conventional Neural Network Let119891119897minus1119895represent the 119895th feature map at the layer 119897minus1The 119895th featuremap at the layer 119897 can be represented as
where119873119897minus119897 represents the number of featuremaps at the 119897minus1thlayer 119896119894119895 represents the kernel function and 119887119897119895 represents thebias at 119897 where 120590 performs a nonlinear function operationThe layer before the Soft-Max Layer can be represented as
Let 119901 = 1 represent Benign class and 119901 = 2 represent theMalignant class The cross-entropy loss of the above functioncan be calculated as
119871119901 = minus ln (119910119901) (10)
Whichever group experiences a large loss value themodel will consider the other group as predicted class
A difficult part of working on DNN is that it requiresa specialized software package for the data analysis Fewresearch groups have been working on how effectively datacan be analyzed by DNN from different perspectives and thedemand Table 8 summarizes some of the software which isavailable for DNN analysis
The history of the CNN and its use for biomedical imageanalysis is a long one Fukushima first introduced a CNNnamed ldquonecognitronrdquo which has the ability to recognizestimulus patterns with a few shifting variances [113] Tothe best of our knowledge Wu et al first classified a setof mammogram images into malignant and benign classesusing a CNN model [78] In their proposed model they onlyutilized one hidden layer After that in 1996 Sahiner et alutilized CNNmodel to classify mass and normal breast tissueand achieved ROC scores of 087 [79] In 2002 Lo et alutilized aMultiple Circular Path CNN (MCPCNN) for tumoridentification from mammogram images and obtained ROCscores of around 089 After an absence of investigation ofthe CNN model this model regained its momentum afterthe work of Krizhevsky et al [114] Their proposed model isknown as AlexNet After this work a revolutionary change
12 Computational and Mathematical Methods in Medicine
Table 8 Available software for deep learning analysis
Software Interface and backend Provider
Caffe [65 66] Python MATLAB C++ Berkeley Vision and Learning CentreUniversity of California Berkeley
Torch [67] C LuaJIT
MatConvNet [68 69] MATLAB C Visual Geometry Group Department ofEngineering University of Oxford
Theano [70 71] Python Montreal Institute for Learning AlgorithmsUniversity of Montreal
TensorFlows [72] C++ Python GoogleCNTK [73] C++ MicrosoftKeras [74] Theano Tensor Flow MITdl4j [75] Java Skymind Engineering
DeeBNET [76 77] MATLAB Information Technology DepartmentAmirkabir University of Technology
has been achieved in the image classification and analysisfield As an advanced engineering of the AlexNet the papertitled ldquoGoing Deeper with Convolutionsrdquo by Szegedy [115]introduced the GoogleNet model This model contains amuch deeper network than AlexNet Sequentially ResNet[116] Inception [117] Inception-v4 Inception-ResNet [118]and a few other models have recently been introduced
Later directly or with some advanced modificationthese DNN models have been adapted for biomedical imageanalysis In 2015 Fonseca et al [81] classified breast densityusing CNN techniques CNN requires a sufficient amountof data to train the system It is always very difficult tofind a sufficient amount of medical data for training a CNNmodel A pretrained CNN model with some fine tuning canbe used rather than create a model from scratch [119] Theauthors of [119] did not perform their experiments on a breastcancer image dataset however they have performed theirexperiments on three different medical datasets with layer-wise training and claimed that ldquoretrained CNN along withadequate training can provide better or at least the sameamount of performancerdquo
The Deep Belief Network (DBN) is another branch of theDeep Neural Network which mainly consists of RestrictedBoltzmann Machine (RBM) techniques The DBN methodwas first utilized for supervised image classification by Liu etal [120] After that Abdel-Zaher and Eldeib utilized the DBNmethod for breast image classification [121] This field is stillnot fully explored for breast image classification yet Zhanget al utilized both RBM and Point-Wise Gated RBM (PRBM)for shear-wave electrography image classification where thedataset contains 227 images [97]Their achieved classificationAccuracy Sensitivity and Specificity are 9340 8860 and9710 respectively Tables 9 10 and 11 have summarized themost recent work for breast image classification along withsome pioneer work on CNN
313 Logic Based Algorithm A Logic Based algorithm isa very popular and effective classification method whichfollows the tree structure principle and logical argument asshown in Figure 16 This algorithm classifies instances based
on the featurersquos values Along with other criteria a decision-tree based algorithm contains the following features
(i) Root node a root node contains no incoming nodeand it may or may not contain any outgoing edge
(ii) Splitting splitting is the process of subdividing a set ofcases into a particular group Normally the followingcriteria are maintained for the splitting
(a) information gain(b) Gini index(c) chi squared
(iii) Decision node(iv) Leafterminal node this kind of node has exactly one
incoming edge and no outgoing edgeThe tree alwaysterminates here with a decision
(v) Pruning pruning is a process of removing subtreesfrom the tree Pruning performs to reduce the over-fitting problem Two kinds of pruning techniques areavailable
(a) prepruning(b) postpruning
Among all the tree based algorithms IterativeDichotomiser 3 (ID3) can be considered as a pioneerproposed by Quinlan [149] The problem of the ID3algorithm is to find the optimal solution which is very muchprone towards overfitting To overcome the limitation of theID3 algorithm the C45 algorithm has been introduced byQuinlan [150] where a pruning method has been introducedto control the overfitting problem Pritom et al [151] classifiedthe Wisconsin breast dataset where they utilized 35 featuresThey have obtained 7630 Accuracy 7510 False PositiveRate and ROC score 0745 when they ranked the featuresWithout ranking the features they obtained 7370Accuracy5070 False Positive Rate and ROC score value 5280 Asriet al [152] utilized the C45 algorithm for the Wisconsin
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
10 Computational and Mathematical Methods in Medicine
Table 7 Neural Network for breast image classification
Reference Descriptor Image type Number ofimages Key findings
Chen et al [61]
(1) Variance Contrast of WaveletCoefficient Ultrasound 242 (1)The achieved ROC curve 09396 plusmn 00183(2) Autocorrelation of WaveletCoefficient
Silva et al [62](1) 22 different morphologicalfeatures such as convexity andlobulation have been utilized
Ultrasound mdash (1)The best obtained Accuracy and ROCcurve are 9698 and 098 respectively
Saritas [63](1) Age of patient (2)massshape (3)mass border (4)Massdensity (5) BIRADS Mammogram mdash
(1) Disease prediction rate is 905(2) Neural Network utilized 5 neurons ininput layers and one hidden layer
Lopez-Melendez etal [64]
(1) Area perimeter etc havebeen utilized Mammogram 322 (1)The achieved Sensitivity and Specificity
are 9629 and 9900 respectively
themodel A kernel of size119898times119899 is scanned through the inputdata for the convolutional operation which ensures the localconnectivity and weight sharing property
(ii) Stride and Padding In the convolutional operation afilter scans through the input matrices In each step howmuch position a kernel filter moves through the matrixis known as the stride By default stride keeps to 1 Withinappropriate selection of the stride the model can lose theborder information To overcome this issue themodel utilizesextra rows and columns at the end of the matrices and theseadded rows and columns contain all 0s This adding of extrarows and columns which contain only zero value is known aszero padding
(iii) Nonlinear Operation The output of each of the kerneloperations is passed through a rectifier function such as Rec-tified Linear Unit (ReLU) Leaky-ReLU TanH and SigmoidThe Sigmoid function can be defined as
120590 (119909) = 1(1 + expminus119909) (3)
and the tanh function can be defined as
tanh (119909) = (exp119909 minus expminus119909)(exp119909 + expminus119909) (4)
However the most effective rectifier is ReLU The ReLUmethod converts all the information into zero if it is less thanor equal to zero and passes all the other data as is shown inFigure 13
120590 (119909) = max (0 119909) (5)
Another important nonlinear function is Leaky-RelU
where 120572 is predetermined parameter which can be varied togive a better model
minus3 minus2 minus1 0 1 2 3
1
2
3
InputO
utpu
t
Figure 13 ReLU Operation
(iv) Subsampling Subsampling is the procedure of reducingthe dimensionality of each of the feature maps of a particularlayer this operation is also known as a pooling operationActually it reduces the amount of feature information fromthe overall data By doing so it reduces the overall computa-tional complexity of themodel To do this 119904times119904 patch units areutilized The two most popular pooling methods are
(a) Max-Pooling
(b) Average Pooling
In Max-Pooling only the maximum values within a partic-ular kernel size are selected for further calculation Consideran example of a 16 times 16 image as shown in Figure 14 A 2 by2 kernel is applied to the whole image 4 blocks in total andproduces a 4 times 4 output image For each block of four valueswe have selected the maximum For instance from blocksone two three and four maximum values 4 40 13 and 8are selected respectively as they are the maximum in thatblock For the Average Pooling operation each kernel givesthe output as average
(v) Dropout Regularization of the weight can reduce theoutfitting problem Randomly removing some neurons can
Computational and Mathematical Methods in Medicine 11
Figure 15 Work-flow of a Convolutional Neural Network
regularize the overfilling problem The technique of ran-domly removing neurons from the network is known asdropout
(vi) Soft-Max Layer This layer contains normalized expo-nential functions to calculate the loss function for the dataclassification
Figure 15 shows a generalized CNN model for the imageclassificationAll the neurons of themost immediate layer of afully connected layer are completely connected with the fullyconnected layer like a conventional Neural Network Let119891119897minus1119895represent the 119895th feature map at the layer 119897minus1The 119895th featuremap at the layer 119897 can be represented as
where119873119897minus119897 represents the number of featuremaps at the 119897minus1thlayer 119896119894119895 represents the kernel function and 119887119897119895 represents thebias at 119897 where 120590 performs a nonlinear function operationThe layer before the Soft-Max Layer can be represented as
Let 119901 = 1 represent Benign class and 119901 = 2 represent theMalignant class The cross-entropy loss of the above functioncan be calculated as
119871119901 = minus ln (119910119901) (10)
Whichever group experiences a large loss value themodel will consider the other group as predicted class
A difficult part of working on DNN is that it requiresa specialized software package for the data analysis Fewresearch groups have been working on how effectively datacan be analyzed by DNN from different perspectives and thedemand Table 8 summarizes some of the software which isavailable for DNN analysis
The history of the CNN and its use for biomedical imageanalysis is a long one Fukushima first introduced a CNNnamed ldquonecognitronrdquo which has the ability to recognizestimulus patterns with a few shifting variances [113] Tothe best of our knowledge Wu et al first classified a setof mammogram images into malignant and benign classesusing a CNN model [78] In their proposed model they onlyutilized one hidden layer After that in 1996 Sahiner et alutilized CNNmodel to classify mass and normal breast tissueand achieved ROC scores of 087 [79] In 2002 Lo et alutilized aMultiple Circular Path CNN (MCPCNN) for tumoridentification from mammogram images and obtained ROCscores of around 089 After an absence of investigation ofthe CNN model this model regained its momentum afterthe work of Krizhevsky et al [114] Their proposed model isknown as AlexNet After this work a revolutionary change
12 Computational and Mathematical Methods in Medicine
Table 8 Available software for deep learning analysis
Software Interface and backend Provider
Caffe [65 66] Python MATLAB C++ Berkeley Vision and Learning CentreUniversity of California Berkeley
Torch [67] C LuaJIT
MatConvNet [68 69] MATLAB C Visual Geometry Group Department ofEngineering University of Oxford
Theano [70 71] Python Montreal Institute for Learning AlgorithmsUniversity of Montreal
TensorFlows [72] C++ Python GoogleCNTK [73] C++ MicrosoftKeras [74] Theano Tensor Flow MITdl4j [75] Java Skymind Engineering
DeeBNET [76 77] MATLAB Information Technology DepartmentAmirkabir University of Technology
has been achieved in the image classification and analysisfield As an advanced engineering of the AlexNet the papertitled ldquoGoing Deeper with Convolutionsrdquo by Szegedy [115]introduced the GoogleNet model This model contains amuch deeper network than AlexNet Sequentially ResNet[116] Inception [117] Inception-v4 Inception-ResNet [118]and a few other models have recently been introduced
Later directly or with some advanced modificationthese DNN models have been adapted for biomedical imageanalysis In 2015 Fonseca et al [81] classified breast densityusing CNN techniques CNN requires a sufficient amountof data to train the system It is always very difficult tofind a sufficient amount of medical data for training a CNNmodel A pretrained CNN model with some fine tuning canbe used rather than create a model from scratch [119] Theauthors of [119] did not perform their experiments on a breastcancer image dataset however they have performed theirexperiments on three different medical datasets with layer-wise training and claimed that ldquoretrained CNN along withadequate training can provide better or at least the sameamount of performancerdquo
The Deep Belief Network (DBN) is another branch of theDeep Neural Network which mainly consists of RestrictedBoltzmann Machine (RBM) techniques The DBN methodwas first utilized for supervised image classification by Liu etal [120] After that Abdel-Zaher and Eldeib utilized the DBNmethod for breast image classification [121] This field is stillnot fully explored for breast image classification yet Zhanget al utilized both RBM and Point-Wise Gated RBM (PRBM)for shear-wave electrography image classification where thedataset contains 227 images [97]Their achieved classificationAccuracy Sensitivity and Specificity are 9340 8860 and9710 respectively Tables 9 10 and 11 have summarized themost recent work for breast image classification along withsome pioneer work on CNN
313 Logic Based Algorithm A Logic Based algorithm isa very popular and effective classification method whichfollows the tree structure principle and logical argument asshown in Figure 16 This algorithm classifies instances based
on the featurersquos values Along with other criteria a decision-tree based algorithm contains the following features
(i) Root node a root node contains no incoming nodeand it may or may not contain any outgoing edge
(ii) Splitting splitting is the process of subdividing a set ofcases into a particular group Normally the followingcriteria are maintained for the splitting
(a) information gain(b) Gini index(c) chi squared
(iii) Decision node(iv) Leafterminal node this kind of node has exactly one
incoming edge and no outgoing edgeThe tree alwaysterminates here with a decision
(v) Pruning pruning is a process of removing subtreesfrom the tree Pruning performs to reduce the over-fitting problem Two kinds of pruning techniques areavailable
(a) prepruning(b) postpruning
Among all the tree based algorithms IterativeDichotomiser 3 (ID3) can be considered as a pioneerproposed by Quinlan [149] The problem of the ID3algorithm is to find the optimal solution which is very muchprone towards overfitting To overcome the limitation of theID3 algorithm the C45 algorithm has been introduced byQuinlan [150] where a pruning method has been introducedto control the overfitting problem Pritom et al [151] classifiedthe Wisconsin breast dataset where they utilized 35 featuresThey have obtained 7630 Accuracy 7510 False PositiveRate and ROC score 0745 when they ranked the featuresWithout ranking the features they obtained 7370Accuracy5070 False Positive Rate and ROC score value 5280 Asriet al [152] utilized the C45 algorithm for the Wisconsin
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Figure 15 Work-flow of a Convolutional Neural Network
regularize the overfilling problem The technique of ran-domly removing neurons from the network is known asdropout
(vi) Soft-Max Layer This layer contains normalized expo-nential functions to calculate the loss function for the dataclassification
Figure 15 shows a generalized CNN model for the imageclassificationAll the neurons of themost immediate layer of afully connected layer are completely connected with the fullyconnected layer like a conventional Neural Network Let119891119897minus1119895represent the 119895th feature map at the layer 119897minus1The 119895th featuremap at the layer 119897 can be represented as
where119873119897minus119897 represents the number of featuremaps at the 119897minus1thlayer 119896119894119895 represents the kernel function and 119887119897119895 represents thebias at 119897 where 120590 performs a nonlinear function operationThe layer before the Soft-Max Layer can be represented as
Let 119901 = 1 represent Benign class and 119901 = 2 represent theMalignant class The cross-entropy loss of the above functioncan be calculated as
119871119901 = minus ln (119910119901) (10)
Whichever group experiences a large loss value themodel will consider the other group as predicted class
A difficult part of working on DNN is that it requiresa specialized software package for the data analysis Fewresearch groups have been working on how effectively datacan be analyzed by DNN from different perspectives and thedemand Table 8 summarizes some of the software which isavailable for DNN analysis
The history of the CNN and its use for biomedical imageanalysis is a long one Fukushima first introduced a CNNnamed ldquonecognitronrdquo which has the ability to recognizestimulus patterns with a few shifting variances [113] Tothe best of our knowledge Wu et al first classified a setof mammogram images into malignant and benign classesusing a CNN model [78] In their proposed model they onlyutilized one hidden layer After that in 1996 Sahiner et alutilized CNNmodel to classify mass and normal breast tissueand achieved ROC scores of 087 [79] In 2002 Lo et alutilized aMultiple Circular Path CNN (MCPCNN) for tumoridentification from mammogram images and obtained ROCscores of around 089 After an absence of investigation ofthe CNN model this model regained its momentum afterthe work of Krizhevsky et al [114] Their proposed model isknown as AlexNet After this work a revolutionary change
12 Computational and Mathematical Methods in Medicine
Table 8 Available software for deep learning analysis
Software Interface and backend Provider
Caffe [65 66] Python MATLAB C++ Berkeley Vision and Learning CentreUniversity of California Berkeley
Torch [67] C LuaJIT
MatConvNet [68 69] MATLAB C Visual Geometry Group Department ofEngineering University of Oxford
Theano [70 71] Python Montreal Institute for Learning AlgorithmsUniversity of Montreal
TensorFlows [72] C++ Python GoogleCNTK [73] C++ MicrosoftKeras [74] Theano Tensor Flow MITdl4j [75] Java Skymind Engineering
DeeBNET [76 77] MATLAB Information Technology DepartmentAmirkabir University of Technology
has been achieved in the image classification and analysisfield As an advanced engineering of the AlexNet the papertitled ldquoGoing Deeper with Convolutionsrdquo by Szegedy [115]introduced the GoogleNet model This model contains amuch deeper network than AlexNet Sequentially ResNet[116] Inception [117] Inception-v4 Inception-ResNet [118]and a few other models have recently been introduced
Later directly or with some advanced modificationthese DNN models have been adapted for biomedical imageanalysis In 2015 Fonseca et al [81] classified breast densityusing CNN techniques CNN requires a sufficient amountof data to train the system It is always very difficult tofind a sufficient amount of medical data for training a CNNmodel A pretrained CNN model with some fine tuning canbe used rather than create a model from scratch [119] Theauthors of [119] did not perform their experiments on a breastcancer image dataset however they have performed theirexperiments on three different medical datasets with layer-wise training and claimed that ldquoretrained CNN along withadequate training can provide better or at least the sameamount of performancerdquo
The Deep Belief Network (DBN) is another branch of theDeep Neural Network which mainly consists of RestrictedBoltzmann Machine (RBM) techniques The DBN methodwas first utilized for supervised image classification by Liu etal [120] After that Abdel-Zaher and Eldeib utilized the DBNmethod for breast image classification [121] This field is stillnot fully explored for breast image classification yet Zhanget al utilized both RBM and Point-Wise Gated RBM (PRBM)for shear-wave electrography image classification where thedataset contains 227 images [97]Their achieved classificationAccuracy Sensitivity and Specificity are 9340 8860 and9710 respectively Tables 9 10 and 11 have summarized themost recent work for breast image classification along withsome pioneer work on CNN
313 Logic Based Algorithm A Logic Based algorithm isa very popular and effective classification method whichfollows the tree structure principle and logical argument asshown in Figure 16 This algorithm classifies instances based
on the featurersquos values Along with other criteria a decision-tree based algorithm contains the following features
(i) Root node a root node contains no incoming nodeand it may or may not contain any outgoing edge
(ii) Splitting splitting is the process of subdividing a set ofcases into a particular group Normally the followingcriteria are maintained for the splitting
(a) information gain(b) Gini index(c) chi squared
(iii) Decision node(iv) Leafterminal node this kind of node has exactly one
incoming edge and no outgoing edgeThe tree alwaysterminates here with a decision
(v) Pruning pruning is a process of removing subtreesfrom the tree Pruning performs to reduce the over-fitting problem Two kinds of pruning techniques areavailable
(a) prepruning(b) postpruning
Among all the tree based algorithms IterativeDichotomiser 3 (ID3) can be considered as a pioneerproposed by Quinlan [149] The problem of the ID3algorithm is to find the optimal solution which is very muchprone towards overfitting To overcome the limitation of theID3 algorithm the C45 algorithm has been introduced byQuinlan [150] where a pruning method has been introducedto control the overfitting problem Pritom et al [151] classifiedthe Wisconsin breast dataset where they utilized 35 featuresThey have obtained 7630 Accuracy 7510 False PositiveRate and ROC score 0745 when they ranked the featuresWithout ranking the features they obtained 7370Accuracy5070 False Positive Rate and ROC score value 5280 Asriet al [152] utilized the C45 algorithm for the Wisconsin
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
12 Computational and Mathematical Methods in Medicine
Table 8 Available software for deep learning analysis
Software Interface and backend Provider
Caffe [65 66] Python MATLAB C++ Berkeley Vision and Learning CentreUniversity of California Berkeley
Torch [67] C LuaJIT
MatConvNet [68 69] MATLAB C Visual Geometry Group Department ofEngineering University of Oxford
Theano [70 71] Python Montreal Institute for Learning AlgorithmsUniversity of Montreal
TensorFlows [72] C++ Python GoogleCNTK [73] C++ MicrosoftKeras [74] Theano Tensor Flow MITdl4j [75] Java Skymind Engineering
DeeBNET [76 77] MATLAB Information Technology DepartmentAmirkabir University of Technology
has been achieved in the image classification and analysisfield As an advanced engineering of the AlexNet the papertitled ldquoGoing Deeper with Convolutionsrdquo by Szegedy [115]introduced the GoogleNet model This model contains amuch deeper network than AlexNet Sequentially ResNet[116] Inception [117] Inception-v4 Inception-ResNet [118]and a few other models have recently been introduced
Later directly or with some advanced modificationthese DNN models have been adapted for biomedical imageanalysis In 2015 Fonseca et al [81] classified breast densityusing CNN techniques CNN requires a sufficient amountof data to train the system It is always very difficult tofind a sufficient amount of medical data for training a CNNmodel A pretrained CNN model with some fine tuning canbe used rather than create a model from scratch [119] Theauthors of [119] did not perform their experiments on a breastcancer image dataset however they have performed theirexperiments on three different medical datasets with layer-wise training and claimed that ldquoretrained CNN along withadequate training can provide better or at least the sameamount of performancerdquo
The Deep Belief Network (DBN) is another branch of theDeep Neural Network which mainly consists of RestrictedBoltzmann Machine (RBM) techniques The DBN methodwas first utilized for supervised image classification by Liu etal [120] After that Abdel-Zaher and Eldeib utilized the DBNmethod for breast image classification [121] This field is stillnot fully explored for breast image classification yet Zhanget al utilized both RBM and Point-Wise Gated RBM (PRBM)for shear-wave electrography image classification where thedataset contains 227 images [97]Their achieved classificationAccuracy Sensitivity and Specificity are 9340 8860 and9710 respectively Tables 9 10 and 11 have summarized themost recent work for breast image classification along withsome pioneer work on CNN
313 Logic Based Algorithm A Logic Based algorithm isa very popular and effective classification method whichfollows the tree structure principle and logical argument asshown in Figure 16 This algorithm classifies instances based
on the featurersquos values Along with other criteria a decision-tree based algorithm contains the following features
(i) Root node a root node contains no incoming nodeand it may or may not contain any outgoing edge
(ii) Splitting splitting is the process of subdividing a set ofcases into a particular group Normally the followingcriteria are maintained for the splitting
(a) information gain(b) Gini index(c) chi squared
(iii) Decision node(iv) Leafterminal node this kind of node has exactly one
incoming edge and no outgoing edgeThe tree alwaysterminates here with a decision
(v) Pruning pruning is a process of removing subtreesfrom the tree Pruning performs to reduce the over-fitting problem Two kinds of pruning techniques areavailable
(a) prepruning(b) postpruning
Among all the tree based algorithms IterativeDichotomiser 3 (ID3) can be considered as a pioneerproposed by Quinlan [149] The problem of the ID3algorithm is to find the optimal solution which is very muchprone towards overfitting To overcome the limitation of theID3 algorithm the C45 algorithm has been introduced byQuinlan [150] where a pruning method has been introducedto control the overfitting problem Pritom et al [151] classifiedthe Wisconsin breast dataset where they utilized 35 featuresThey have obtained 7630 Accuracy 7510 False PositiveRate and ROC score 0745 when they ranked the featuresWithout ranking the features they obtained 7370Accuracy5070 False Positive Rate and ROC score value 5280 Asriet al [152] utilized the C45 algorithm for the Wisconsin
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Computational and Mathematical Methods in Medicine 13
Table 9 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Wu et al [78] (1) Global Features Mammogram 40 (1) Achieved Sensitivity 7500 and Specificity7500
Sahiner et al [79] (1) Global Features Mammogram 168 (1)The achieved ROC score is 087
Lo et al [80] (1) Density size ShapeMargin Mammogram 144 (1)The achieved ROC curve is 089
Fonseca et al [81] (1) Global Features Mammogram mdash(1) Breast density classification has beenperformed utilizing HT-L3 convolution(2)Average achieved obtained Kappa value is 058
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The achieved ROC curve is 0826
Su et al [83] (1) Global Features Mammogram 92(1) Fast Scanning CNN (fCNN) method has beenutilized to reduce the information loss(2)The average Precision Recall and 1198651 score are9100 8200 and 085 respectively
Sharma and Preet [84] (1) GLCM GLDMGeometrical Mammogram 40
(1)The best Accuracy achieved is 7523 and7234 respectively for fatty and dense tissueclassification
Spanhol et al [6] (1) Global Features Histopathology 7909 (1)The best Accuracy achieved 89 plusmn 66
Rezaeilouyeh et al [85] (1) Local and GlobalFeatures Histopathology mdash
(1) Shearlet transform has been utilized forextracting local features(2)When they utilize RGB image along withmagnitude of Shearlet transform together theAchieved Sensitivity Specificity and Accuracywere 8400 plusmn 100 9100 plusmn 200 and 8400 plusmn400 when they utilize RGB image along withboth the phase and magnitude of Shearlettransform together the achieved SensitivitySpecificity and Accuracy were 8900 plusmn 1009400 plusmn 100 and 8800 plusmn 500
Root node
Decision node Decision node
Decision node Terminalnode
Terminalnode
Terminalnode
Terminalnode
Terminalnode
Node split
Subtree
Figure 16 A general structure of a tree
database classification where they utilized 11 features andobtained 9113 Accuracy
Logic Based algorithms allow us to produce more thanone tree and combine the decisions of those trees for anadvanced result this mechanism is known as an ensemblemethod An ensemble method combines more than one
classifier hypothesis together and produces more reliableresults through a voting concept Boosting and baggingare two well-known ensemble methods Both boosting andbagging aggregate the trees The difference is in baggingsuccessive trees do not depend on the predecessor treeswhere in the boosting method successive trees depend on the
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
14 Computational and Mathematical Methods in Medicine
Table 10 Convolutional Neural Network
Reference Descriptor Image type Number ofimages Key findings
Albayrak and Bilgin [86] (1) Global Features Histopathology 100
(1) Cluster-based segmentation has beenperformed to find out the cellular structure(2) Blob analysis has been performed on thesegmented images(3) To reduce the high dimensionality PrincipalComponent Analysis (PCA) and LinearDiscriminant Analysis (LDA) methods have beenutilized(4) Before the dimensionality reduction thePrecision Recall and 119865-score values were 97206600 and 078 respectively but when thedimensionality reduction method was utilized thePrecision Recall and 119865-score values were10000 9400 and 096 respectively(5)The best average Accuracy is 7300 (withoutdimensionality reduction) and 968 (withdimensionality reduction)
Jiao et al [87] (1) Global and LocalFeatures Mammogram mdash
(1)They performed their experiments on theDDSM database(2) Total required parameter is 58 times 107 and timefor the per image processing is 110 ms(3)The best classification achieved is 9670however they show that when they utilize theVGG model the Accuracy was 9700 which isslightly better than their modelHowever in terms of memory size and time perimage processing their model gives betterperformance than the VGG model
Zejmo et al [88] (1) Global Features Cytology 40
(1) GoogleNet and AlexNet models have beenutilized(2)The best Accuracy obtained when they utilizedGoogleNet model was 8300
information gathered from the predecessor trees Gradientboosting is a very popular method for data classification[153 154] however a state-of-the-art boosting algorithm suchas ldquoExtreme Gradient Boostingrdquo (XGBoosting) is a veryeffective method for data classification [155] Interestinglythere has not been a single paper published for breast imageclassification using the XGBoost algorithm Along with theboosting method different bagging methods are availableamong them Random Forest (RF) is very popular where alarge number of uncorrelated trees are aggregated togetherfor a better prediction Tables 12 and 13 summarize a set ofpapers where a Logic Based algorithm has been used forimage classification
314 Support Vector Machine (SVM) SVM were proposedby VC (Vepnick-Cherovorenkis) This technique does notrequire any prior distribution knowledge for the data classi-fication task like Bayesian classification technique In manypractical situations the distribution of the features is notavailable In such cases SVM can be used to classify theavailable data into the different classes
Consider the set of two-dimensional data plotted inFigure 17The symbol ldquo∘rdquo represents those data which belongto Class-1 and ldquo◻rdquo represents data which belong to Class-2A hyperplane (119875) has been drawn which classifies the datainto two classes Interestingly there will be ldquo119899rdquo hyperplanesavailable which can separate the data
Let X = X119894 where X119894 isin R119899 (119894 = 1 2 3 119897) isto be classified into two classes 120596 isin 1205961 1205962 Suppose thatthe classes 1205961 and 1205962 are recognized as ldquo+1rdquo and ldquominus1rdquoClassification of this data can be written
During the learning stage the SVM finds parameters W119894 =[1198821119894 1198822119894 119882119899119894 ]119879 and 119887 to produce a decision function119889(X119894W119894 119887)119889 (X119894W119894 119887) = W119879119894 X119894 + 119887 = W119894 sdot X119894 + 119887
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Computational and Mathematical Methods in Medicine 15
Table 11 Convolutional Neural Network
Reference Descriptor Image type Number of images Key findings
Jiang et al [89] (1) Global Features Mammogram mdash
(1) Image preprocessing was performed toenhance tissue characteristics(2) Transfer learning was performed and obtainedAUC was 088 whereas when the system learnedfrom scratch the best ROC is 082
Suzuki et al [90] (1) Global Features Mammogram 198 (1)The achieved sensitivity 8990(2) Transfer learning techniques have beenutilized
Qiu et al [91] (1) Global Features Mammogram 270 (1) Average achieved Accuracy is 7140
Samala et al [92] (1) Global Features mdash 92(1)They utilized Deep Learning CNN (DLCNN)and CNNmodels for classification(2)The AUC of CNN and DLCNNmodel is 089and 093 respectively
Sharma and Preet [84] (1) Global Features Mammogram 607
(1) Transfer learning and ensemble techniquesutilized(2)When using ensemble techniques the softvoting method has been used(3)The best ROC score is 086
Kooi et al [93] (1) Global and Localfeatures Mammogram 44090 (1) Transfer learning method utilized (VGG
model)
Geras et al [94] (1) Global Features Mammogram 102800 (1)They investigated the relation of the Accuracywith the database size and image size
Arevalo et al [82] (1) Global Features Mammogram 736 (1)The best ROC value was 0822
Table 12 Logic Based
Reference Descriptor Image type Numberof images Key findings
Beura et al [95]
(1) Two-dimensionaldiscrete orthonormal119878-transform has been usedfor the feature extraction
Mammogram mdash
(1) Achieved Accuracy and AUC values on MIASdatabase are 983 09985(2) Achieved Accuracy and AUC values onDDSM database are 988 09992
Diz et al [96] (1) GLCM Mammogram 410 (1)Their achieved Accuracy value is 7660(2) GLRLM (2)Mean false positive value is 8100
Zhang et al [97] (1) 133 features (mass basedand content based) Mammogram 400
(1) Computer model has been created which isable to find a location that was not detected bytrainee
Ahmad and Yusoff[98] (1) Nine features selected Biopsy 700 (1) Achieved Sensitivity Specificity and Accuracy
are 7500 7000 and 7200 respectively
Paul et al [99] (1)Harlick texture feature Histopathological 50 (1)Their achieved Recall and Precision are 8113and 8350
Chen et al [100]
(1) Dual-tree complexwavelet transform(DT-CWT) has been usedfor the feature extraction
Mammogram mdash (1) Achieved Received Operating Curve (ROC)0764
Zhang et al [101] (1) Curvelet Transform(2) GLCM (3) CLBP Histopathological 50
(1) Random Subspace Ensemble (RSE) utilized(2)Their achieved classification Accuracy is9522 where the previous Accuracy on this samedatabase was 9340
16 Computational and Mathematical Methods in Medicine
Table 13 Logic Based
Reference Descriptor Image type Numberof images Key findings
Angayarkanni andKamal [102] (1) GLCM Mammogram 322 (1)The Achieved Sensitivity and Accuracy are 9340
and 9950 respectively
Wang et al [103]
(1)Horizontal WeightedSum(2) Vertical Weighted Sum(3) Diagonal WeightedSum(4) Grid Weighted Sum
Mammogram 322
(1) Surrounding Region Dependence Method (SRDM)utilized for region detection(2) Achieved True Positive Rate 9000 and FalsePositive Rate 8880
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
(1) ANOVA method utilized for feature prioritization(2)When they use RF algorithm on Mammogram(DDSM) dataset obtained Accuracy and ROC are7900 and 089
Muramatsu et al[105]
(1) Radial Local TernaryPattern (RLTP) Mammogram 376
(1) Textural features have been extracted from theregions of interest (ROIs) using RLTP(2)They claimed that the RLTP feature provides betterperformance than the rotation invariant patterns
Dong et al [106](1) NRL margin gradient(2) Gray-level histogram(3) Pixel value fluctuation Mammogram mdash
(1) Chain code utilized for extraction of regions ofinterest (ROIs)(2) Rough-Set method utilized to enhance the ROIs(3)Their achieved ROC value is 0947 and obtainedMatthews Correlation (MCC) is 08652
Piantadosi et al[107]
(1) Local BinaryPattern-Three OrthogonalProjections (LBP-TOP)
Mammogram mdash (1)Their achieved Accuracy Sensitivity and Specificityvalues are 8460 8000 and 9090
X
Y
Hyperplane P
Figure 17 SVM finds the hyperplane which separates two classes
whereW119894X119894 isin R119899 As the training data are linearly separableno training data will satisfy the condition
119889 (X119894W119894 119887) = 0 (13)
To control the separability we consider the followinginequalities
119889 (X119894W119894 119887) ge 1 for 120596119894 = +1119889 (X119894W119894 119887) lt 1 for 120596119894 = minus1 (14)
Sometime it is very difficult to find the perfect hyperplanewhich can separate the data but if we transform the datainto a higher dimension the data may be easily separableTo separate this kind of data a kernel function can beintroduced
Kernel Methods Assume a transformation 120601 such that ittransforms the dataset X1 isin R119899 into dataset X2 isin R119898 where119898 gt 119899 Now train the linear SVM on the dataset X2 to get anew classifier 119865SVM
A kernel 120601 effectively computes a dot product in a higher-dimensional space R119898 For x119894 x119895 isin R119873 119870(x119894 x119895) =⟨120601(x119894 x119895)⟩119898 is an inner product ofR119898 where120601(x) transformsx to R119898 Consider x119894 x119895 isin R119899 then we can define thekernel as follows
(i) Radial basis function kernel (rbf) 119870(x119894 x119895) =exp(minus120574| lt 120601(x119894 minus x119895) gt |2)
(iii) Sigmoid kernel119870(x119894 x119895) = tanh(⟨120601(x119894 x119895)⟩ + 119903)(iv) Linear kernel (linear) 119870(x119894 x119895) = ⟨120601(x119894 x119895)⟩The advantage of the kernel method for breast cancer
image classification using an SVM was first introduced byEl-Naqa et al [156] They classify Microcalcification clustersin mammogram images (76 images were utilized for the
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Computational and Mathematical Methods in Medicine 17
Table 14 SVM for breast image classification (Page-1)
Reference Descriptor Image type Numberof images Key findings
Malik et al [108](1) Speed of sound(2) Attenuation image vector(3) Reflection image vector
QTUS mdash
(1) Glands fat skin and connective tissue havebeen classified(2) Both linear and nonlinear SVM classifier havebeen utilized(3)Their experiment obtained 8520 Accuracy
Chang et al [109]
(1) Textural features such as(i) AutocorrelationCoefficient(ii) AutocovarianceCoefficient
Ultrasound 250
(1) Benign and malignant images have beenclassified(2) Accuracy Sensitivity Specificity positivepredictive values and negative predictive valueare 8560 9545 7786 7721 and 9561respectively
Akbay et al [110] (1) 52 features have beenextracted Mammogram mdash (1)Microcalcification (MC) Classification
Accuracy 9400
Levman et al [111]
(1) Relative SignalIntensities(2) Derivative of SignalIntensities(3) Relative Signal Intensitiesand their derivatives in onevector(4) (i) Maximum of signalintensity enhancement (ii)time of maximumenhancement (iii) time ofmaximum washout
MRI 76
(1) Benign and malignant lesions are investigated(2) Linear kernel a polynomial kernel and aradial basis function kernel utilized along with theSVMmethod for the breast image classification
de OliveiraMartins et al[112]
(1) Ripleyrsquos 119870 function Mammogram 390
(1) Benign and malignant image classification(2)The achieved Accuracy Sensitivity andSpecificity are 9494 9286 and 9333respectively
experiment where the total number of MCs was 1120) Theyutilized the SVM method along with the Gaussian kernelas well as the polynomial kernel In 2003 Chang et alclassified a set of sonography images using SVM techniqueswhere they consider that the image is surrounded by picklenoise [157] where the database contains 250 images Theirachieved Accuracy was 9320 A total of thirteen featuresincluding shape law and gradient features were utilizedalong with SVM and a Gaussian kernel for the mammogramimage classification They performed their operation on 193mammogram images and achieved 8370 sensitivity and3020 False Positive Rate [158] SVM has been combinedwith the NN method by B Sing et al for ultrasound breastimage classification where the database contained a totalof 178 images They performed a hybrid feature selectionmethod to select the best features [159]
A breast ultrasound image is always very complex innature The Multiple Instance Learning (MIL) algorithm hasbeen first used along with SVM for the breast image classi-fication by [176] and their obtained Accuracy was 9107The Concentric Circle BOW feature extraction method wasutilized to extract the features and later the SVM methodwas used for breast image classification [177] Their achievedAccuracy is 8833 when the dimension of the features was
1000 A Bag of Features has been extracted from histopatho-logical images (using SIFT and DCT) and using SVM forclassification by Mhala and Bhandari [178] The experimentis performed on a database which contains 361 images where119 images are normal 102 images are ductal carcinomain situ and the rest of the images are invasive carcinomaTheir experiment achieved 10000 classification Accuracyfor ductal carcinoma in situ 9888 classification Accuracyfor invasive carcinoma and 10000 classification Accuracyfor normal image classification A mammogram (DDSM)image database has been classified byHiba et al [179] by SVMalong with the Bag of Feature method Firstly the authorsextract LBP and quantize the binary pattern information forfeature extraction Their obtained Accuracy was 9125
Along with the above-mentioned work different breastimage databases have been analyzed and classified usingSVMWe have summarized some of the work related to SVMin Tables 14 15 and 16
315 Bayesian A Bayesian classifier is a statistical methodbased on Bayes theorem This method does not follow anyexplicit decision rule however it depends on estimatingprobabilitiesThe Naive Bayes method can be considered oneof the earlier Bayesian learning algorithms
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
18 Computational and Mathematical Methods in Medicine
Table 15 SVM for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhang et al [122](1) Fractional Fouriertransform informationutilized as features
Mammogram 200
(1)They selected ROI for avoiding redundant complexity(2)When SVM and Principal Component Analysis wereused together the achieved Accuracy Sensitivity andSpecificity are 9216 plusmn 360 9210 plusmn 275 and9222 plusmn 416 respectively
Shirazi and Rashedi[123] (1) GLCM Ultrasound 322
(1) ROI extracted for reducing redundant complexity(2) SVM and Mixed Gravitational Search Algorithm(MGSA) used together for feature reduction(3)The achieved Accuracy 8600 however SVM withMGSA method achieved 9310 Accuracy
Reference Descriptor Image type Numberof images Key findings
Taheri et al [126](1) Intensity information(2) Value of detected corner(3) Energy Mammogram 600
(1) Classified images into normal and abnormalimages(2) Removing unwanted objects from the images forreducing the redundancy and computationalcomplexity(3) Achieved Precision and Recall rates are 9680and 925 respectively
Tan et al [127]
(1) Shape fat presence ofcalcification texturespiculation ContrastIsodensity type featuresselected(2) Total number of features181
Mammogram 1200
(1) Features have been selected from the region ofinterest(2)They utilized the radial basis function (RBF) fortheir analysis(3)The Sequential Forward Floating Selection(SFFS) method utilized for the feature selection(4)The area under the receiver operatingcharacteristic curve was (AUC) = 0805 plusmn 0012
Kavitha andThyagharajan [128]
(1)Histogram of the intensityhas been used as a statisticalfeature(2) 2D Gabor filter utilized forthe textural feature extraction(3) Clinical features extractedfrom the database directly
Mammogram 322
(1)When using SVM with the linear kernel theobtained Accuracy Sensitivity and Specificity are98 100 and 96 respectively(2)When using weighted feature SVM with weightsthe obtained Accuracy Sensitivity and Specificity are90 100 and 75 respectively
The Naive Bayes (NB) method works on the basis of theBayes formula where each of the features is considered statis-tically independent Consider a dataset with119898 samples witheach sample containing a feature vector xk with 119899 features[180] and belonging to a particular class 119888119896 According to theNB formula the probability of the particular class 119888119896 with theconditional vector xk is represented as
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
(1) Locality Sensitive Discriminant Analysis (LSDA) forthe data reduction(2) NB obtained 8434 Accuracy and 8369Sensitivity with 9086 Specificity
Perez et al [133] (1) 23 features utilized Mammogram mdash (1) UFilter feature selection methods utilized and itsefficiency verified by Wilcoxon statistical test
Rashmi et al [134] (1) 10 features utilized mdash mdash (1) Benign and malignant tumors have been classified
Gatuha and Jiang[135] (1) 10 features utilized mdash mdash
(1)They built an android based benign and malignanttumor classifier(2)Their obtained Accuracy is 964
The NB method is very easy to construct and very firstto predict the data This method can also utilize the kernelmethod However for a large dataset and continuous datathis method has very poor performance NB can be classifiedinto the following subclasses
One of the constraints of the NB classifier is that itconsiders that all the features are conditionally independentA Bayesian Network is another Bayesian classifier whichcan overcome this constraint [181 182] The literature showsthat the Bayesian classifier method is not utilized much forbreast image classification In 2003 Butler et al used NBclassifier for X-ray breast image classification [183] Theyextracted features from the low-level pixels For all featurecombinations they obtained more than 9000 AccuracyBayesian structural learning has been utilized for a breastlesion classifier by Fischer et al [184] Soria et al [185] classifya breast cancer dataset utilizing C45 multilayered percep-tron and the NB algorithm using WEKA software [186]They conclude that the NB method gives better performancethan the other two methods in that particular case Theyalso compared their results with the Bayes classifier outputSome other research on the Bayes classifier and breast imageclassification has been summarized in Tables 17 and 18
32 Performance Based on Unsupervised Learning Thislearning algorithm does not require any prior knowledgeabout the target The main goal of the unsupervised learningis to find the hidden structure and relations between the
different data [187] and distribute the data into differentclusters Basically clustering is a statistical process where aset of data points is partitioned into a set of groups knownas a cluster The119870-means algorithm is a clustering algorithmproposed by [188] Interestingly unsupervised learning canbe utilized as preprocessing step too
(i) In the 119870-means algorithm firstly assign 119870 centroidpoints Suppose that we have 119899 feature points 119909119894where 119894 isin 1 119899 The objective of the 119870-meansalgorithm is to find positions 120583119894 where 119894 isin 1 119870that minimize the data points to the cluster by solving
1003817100381710038171003817119909 minus 12058311989410038171003817100381710038172 (18)
(ii) Self-OrganizingMap (SOM) SOM is another popularunsupervised classifier proposed by Kohonen et al[189ndash191] The main idea of the SOM method is toreduce the dimension of the data and represent thosedimensionally reduced data by a map architecturewhich provides more visual information
(iii) Fuzzy 119862-Means Clustering (FCM) the FCM algo-rithm cluster databased on the value of a member-ship function is proposed by [192] and improved byBezdek [193]
The history of using unsupervised learning for breastimage classification is a long one In 2000 Cahoon et al [194]classified mammogram breast images (DDSM database) inan unsupervised manner utilizing the 119870-NN clustering andFuzzy 119862-Means (FCM) methods Chen et al classified a setof breast images into benign and malignant classes [164]
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
20 Computational and Mathematical Methods in Medicine
Table 18 Bayesian classifier
Reference Descriptor Image type Numberof images Key findings
Benndorf et al [136] (1) BI-RADS featuresutilized mdash 2766
(1) For the training data the AUC value is 0959 for theinclusive model whereas AUC value is 0910 for thedescriptor model
Rodrıguez-Lopezand Cruz-Barbosa[137]
(1) Eight imagefeature nodes utilized mdash mdash (1) NB model obtained 7900 Accuracy 8000
Sensitivity
Nugroho et al [138] (1) Eight imagefeature nodes utilized Mammogram mdash
(1) Naive Bayes model along with SMO obtained ROCvalue is 0903(2) Bayesian Network model along with SMO obtainedAccuracy was 8368
Rodrıguez-Lopezand Cruz-Barbosa[139]
(1) Eight imagefeatures have beenutilized
mdash 231(1) Bayesian Network model obtained 8200Accuracy 8000 Sensitivity and 8300 Specificitywhen they utilized only three features
Shivakumari et al[140] mdash 231
(1) Analyze the Ljubljana breast image dataset(2) NB algorithm along with feature rankingtechniques the best achieved Accuracy was 8146
Rodrıguez-Lopezand Cruz-Barbosa[141]
(1) Seven differentclinical featuresextracted
Mammogram 690 (1) Obtained Accuracy Sensitivity and Specificity are8200 8000 and 8300 respectively
Table 19 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image type Numberof images Key findings
Moftah et al [142] (1) Intensity distributionused as feature MRI mdash
(1)Three types of evaluation measures performed(a) Accuracy (b) feature based (c) shape basedmeasure(2)This can classify the data as well as identify thetarget(3)The obtained best Accuracy of the segmented ROI is9083
Lee et al [143] (1) 1734 signal patterns MRI 322 (1) Available signal patterns have been classified into 10classes
Dalmiya et al [144] (1) Discrete WaveletTransform Mammogram mdash (1) Cancer tumor masses have been segmented
Elmoufidi et al [145] (1) Local Binary Pattern Mammogram 322
(1) Image enhancing(2) Generation of number of clusters(3) Detection of regions of interest(4)Mean detection of regions of interest is 8500
Samundeeswariet al [146] Ultrasound mdash
(1) Utilizing ant colony and regularization parameters(2)This method obtained 9600 similarity betweensegmented and reference tumors
(1) Early detection of tumors from the breast image(2) Tumor detection Accuracy 9232 Sensitivity9024
Chandra et al [148] (1) Gray intensity values Mammogram mdash (1)Mammogram image has been clustered using SOMalong with the Quadratic Neural Network
They utilized a SOM procedure to perform this classificationoperationThey collected 24 autocorrelation textural featuresand used a 10-fold validation method Markey et al utilizedthe SOM method for BIRADS image classification of 4435samples [195] Tables 19 and 20 summarize the breast imageclassification performance based on 119870-means algorithm andSOMmethod
33 Performance Based on Semisupervisor Theworking prin-ciple of semisupervised learning lies in between supervisedand unsupervised learning For the semisupervised learninga few input data have an associated target and large amountsof data are not labeled [196] It is always very difficult to collectthe labeled data Few data such as speech or informationscratched from the web are difficult to label To classify
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Computational and Mathematical Methods in Medicine 21
Table 20 119870-means Cluster Algorithm and Self-Organizing Map for breast image classification
Reference Descriptor Image Type No ofImages Key Findings
Lashkari andFirouzmand[160]
Thermogram 23
(1) Both FCMmethod and Adaboost methodutilized separately to classify images(2) For the classification purposes selected 23features and also select the best features usingfeature selection algorithm When they used theFCMmethod the obtained Mean Accuracy was7500 whereas the Adaboost method Accuracywas 8800
Nattkemper et al[161] MRI mdash (1) 119870-means algorithm as well as SM method
utilizedSlazar-Licea et al[162] sdot sdot sdot mdash (1) Fuzzy 119888-means algorithm used
Marcomini et al[163]
(1) 24 morphologicalfeatures Ultrasound 144
(1)Minimizing noise using Wiener filterequalized and Median filter(2) Obtained Sensitivity 100 and Specificity7800
Chen et al [164] (1) 24 autocorrelationtexture features Ultrasound 243 (1)Obtained ROC area 09357 plusmn 00152 Accuracy
Ultrasound mdash (1) Automated threshold scheme introduce toincrease the robustness of the SOM algorithm
this kind of data semisupervised learning is very efficientHowever lately this method has been utilized for the bratsimage classification too Semisupervised learning can beclassified as
(i) Graph Based (GB)(ii) Semisupervised Support Vector Machine(iii) Human Semisupervised Learning
To the best of our knowledge Li and Yuen have utilized GBsemisupervised learning for biomedical image classification[197] The kernel trick is applied along with the semisu-pervised learning method for breast image classification byLi et al [198] They performed their experiments on theWisconsin Prognostic Breast Cancer (WPBC) dataset forthe breast image classification Ngadi et al utilized both theSKDA (Supervised Kernel-Based Deterministic Annealing)and NSVC methods for mammographic image classification[199] They performed their experiments on 961 imageswhere 5360 of the images were benign and the rest of theimages are malignant Among the other utilized features theyutilized BI-RADS descriptors as features When they utilizedthe NSVC method they also utilized RBF polynomial andlinear kernel They found that the best Accuracy of 9927was achieved when they utilized linear kernels Few studieshave performed the breast image classification by semisuper-vised learning as summarized in Tables 21 and 22
4 Conclusion
Breast cancer is a serious threat to women throughout theworld and is responsible for increasing the female mortality
rate The improvement of the current situation with breastcancer is a big concern and can be achieved by properinvestigation diagnosis and appropriate patient and clinicalmanagement Identification of breast cancer in the earlierstages and a regular check of the cancer can save many livesThe status of cancer changes with time as the appearancedistribution and structural geometry of the cells are changingon a particular time basis because of the chemical changeswhich are always going on inside the cellThe changing struc-ture of cells can be detected by analysing biomedical imageswhich can be obtained by mammogram MRI and so forthtechniques However these images are complex in nature andrequire expert knowledge to perfectly analyze malignancyDue to the nontrivial nature of the images the physiciansometimes makes a decision which might contradict othersHowever computer-aided-diagnosis techniques emphasisingthe machine learning can glean a significant amount ofinformation from the images and provide a decision basedon the gained information such as cancer identification byclassifying the images
The contribution of machine learning techniques toimage classification is a long story Using some advancedengineering techniques with somemodifications the existingmachine learning based image classification techniques havebeen used for biomedical image classification specially forbreast image classification and segmentation A few branchesof the machine learning based image classifier are availablesuch as DeepNeural Network Logic Based and SVM Exceptfor deep-learning a machine learning-based classifier largelydepends on handcrafted feature extraction techniques such asstatistical and structural information that depend on variousmathematical formulations and theorize where they gain
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
22 Computational and Mathematical Methods in Medicine
Table 21 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key finding
Cordeiro et al[166]
(1) Zernikemoments have beenused for the featureextraction
mdash 685(1) Semisupervised Fuzzy GrowCut algorithm utilized(2) For the fatty-tissue classification this methodachieved 9128 Accuracy
Cordeiro et al[167] mdash Mammogram 322
(1) Semisupervised Fuzzy GrowCut as well as the FuzzyGrowCut algorithm utilized for tumors regionsegmentation
Nawel et al [168] mdash mdash mdash
(1) Semisupervised Support Vector Machine (S3VM)utilized(2)This experiment shows impressive results on theDDSM database
Zemmal et al [169] mdash DDSM mdash(1) Transductive semisupervised learning techniqueusing (TSVM) utilized for classification along withdifferent features
Zemmal et al [170] mdash mdash 200 (1) Semisupervised Support Vector Machine (S3VM)utilized with various kernels
Zemmal et al [171](1) GLCM (2)Humoments (3)Central Moments
Mammogram mdash
(1) Transductive Semisupervised learning techniqueused for image classification(2)This experiment shows impressive results on DDSMdatabase
Histopathological 322(1)The Ordering Points to Identify the ClusteringStructure (OPTICS) method utilized for imageclassification [173]
Table 22 Semisupervised algorithm for breast image classification
Reference Descriptor Image type Numberof images Key findings
Zhu et al [174](1) Relative local intensity(2) Shape irregularity(3) Orientation consistency
Ultrasound 144(1) One important microenvironment inside thetumor is vasculature which has been classified inthis paper
Liu et al [175] mdash Ultrasound mdash
(1) Iterated Laplacian regularization basedsemisupervised algorithm for robust featureselection (Iter-LR-CRFS) utilized(2)The archived Accuracy and Sensitivity are890 plusmn 36 and 910 plusmn 52
object-specific information They are further utilized as aninput for an image classifier such as SVM and Logic Basedfor the image classification
This investigation finds that most of the conventionalclassifiers depend on prerequisite local feature extractionThenature of cancer is always changing so the dependencieson a set of local features will not provide good results ona new dataset However the state-of-the art Deep NeuralNetworks specially CNN have recently advanced biomedicalimage classification due to the Global Feature extractioncapabilities As the core of the CNN model is the kernelwhich gives this model the luxury of working with the GlobalFeatures these globally extracted features allow the CNNmodel to extract more hidden structure from the imagesThis allows some exceptional results for breast cancer imageclassification As the CNN model is based on the Global
Features this kind of classifier model should be easy to adaptto a new dataset
This paper also finds that the malignancy information isconcentrated in the particular area defined as ROI Utiliz-ing only the ROI portions information gathered from thesegmented part of the data can improve the performancesubstantially The recent development of the Deep NeuralNetwork can also be utilized for finding the ROI andsegmenting the data which can be further utilized for theimage classification
For breast cancer patient care the machine learning tech-niques and tools have been a tremendous success so far andthis success has gained an extra impetus with the involvementof deep-learning techniques However the main difficulty ofhandling the current deep-learning based machine learningclassifier is its computational complexity which is much
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Computational and Mathematical Methods in Medicine 23
higher than for the traditional method The current researchis focused on the development of the light DNN model sothat both the computational and timing complexities can bereduced Another difficulty of using the DNN based cancerimage classifier is that it requires a large amount of trainingdata However the reinforcement of learning techniques anddata augmentation has been largely adapted with the currentCNN model which can provide reliable outcomes Ourresearch finds that the current trend of machine learningis largely towards deep-learning techniques Among a fewother implications the appropriate tools for designing theoverall deep-learning model was the initial obligation forutilizing deep-learning based machine learning techniquesHowever some reliable software has been introduced whichcan be utilized for breast image classification Initially it wasdifficult to implement a DNN based architecture in simplerdevices however due to cloud-computer based ArtificialIntelligence techniques this issue has been overcome andDNN has already been integrated with electronic devicessuch as mobile phones In future combining the DNNnetwork with the other learning techniques can providemore-positive predictions about breast cancer
Due to the tremendous concern about breast cancermany research contributions have been published so farIt is quite difficult to summarize all the research workrelated to breast cancer image classification based onmachinelearning techniques in a single research article Howeverthis paper has attempted to provide a holistic approachto the breast cancer image classification procedure whichsummarizes the available breast dataset generalized imageclassification techniques feature extraction and reductiontechniques performance measuring criteria and state-of-the-art findings
In a nutshell the involvement of machine learning forbreast image classification allows doctors and physicians totake a second opinion and it provides satisfaction to andraises the confidence level of the patient There is also ascarcity of expert people who can provide the appropriateopinion about the disease Sometimes the patient might needto spend a long time waiting due to the lack of expertpeople In this particular scenario themachine learning baseddiagnostic system can help the patient to receive the timelyfeedback about the disease which can improve the patient-management scenario
Conflicts of Interest
The authors declare that there are no conflicts of interestregarding the publication of this paper
References
[1] J E Skandalakis Embryology and Anatomy of the BreastSpringer BerlinHeidelberg Berlin Heidelberg Germany 2009
[2] M A Shampo and R A Kyle ldquoKarl theodore dussikmdashpioneerin ultrasoundrdquo Mayo Clinic proceedings vol 70 no 12 p 11361995
[3] O H Karatas and E Toy ldquoThree-dimensional imaging tech-niques a literature reviewrdquo European Journal of Dentistry vol8 no 1 pp 132ndash140 2014
[4] M Lakrimi AMThomas G Hutton et al ldquoThe principles andevolution of magnetic resonance imagingrdquo Journal of PhysicsConference Series vol 286 no 1 Article ID 012016 2011
[5] httpwwwaihwgovauacim-books[6] F A Spanhol L S Oliveira C Petitjean and L Heutte ldquoBreast
cancer histopathological image classification using Convolu-tional Neural Networksrdquo inProceedings of the 2016 InternationalJoint Conference on Neural Networks IJCNN 2016 pp 2560ndash2567 Canada July 2016
[7] R M Haralick ldquoStatistical and structural approaches to tex-turerdquo Proceedings of the IEEE vol 67 no 5 pp 786ndash804 1979
[8] H Tamura S Mori and T Yamawaki ldquoTextural features corre-sponding to visual perceptionrdquo IEEE Transactions on SystemsMan and Cybernetics vol 8 no 6 pp 460ndash473 1978
[9] T Lindeberg ldquoFeature detectionwith automatic scale selectionrdquoInternational Journal of Computer Vision vol 30 no 2 pp 79ndash116 1998
[10] C Harris and M Stephens ldquoA combined corner and edgedetectorrdquo in Proceedings of the 4th Alvey Vision Conference pp147ndash151 1988
[11] S M Smith and J M Brady ldquoSUSAN a new approach tolow level image processingrdquo International Journal of ComputerVision vol 23 no 1 pp 45ndash78 1997
[12] E Rosten and T Drummond ldquoFusing points and lines for highperformance trackingrdquo in Proceedings of the 10th IEEE Inter-national Conference on Computer Vision (ICCV rsquo05) vol 2 pp1508ndash1515 Beijing China October 2005
[13] E Rosten andTDrummond ldquoMachine learning for high-speedcorner detectionrdquoProceedings of the 9th EuropeanConference onComputer Vision (ECCV rsquo06) vol Part I Springer-Verlag pp430ndash443 2006
[14] R Lenz ldquoRotation-invariant operators and scale-space filter-ingrdquo Pattern Recognition Letters vol 6 no 3 pp 151ndash154 1987
[15] R Lakemond S Sridharan and C Fookes ldquoHessian-basedaffine adaptation of salient local image featuresrdquo Journal ofMathematical Imaging and Vision vol 44 no 2 pp 150ndash1672012
[16] T Lindeberg ldquoScale selection properties of generalized scale-space interest point detectorsrdquo Journal of Mathematical Imagingand Vision vol 46 no 2 pp 177ndash210 2013
[17] D G Lowe ldquoDistinctive image features from scale-invariantkeypointsrdquo International Journal of Computer Vision vol 60 no2 pp 91ndash110 2004
[18] W N J Hj Wan Yussof and M S Hitam ldquoInvariant Gabor-based interest points detector under geometric transformationrdquoDigital Signal Processing vol 25 no 1 pp 190ndash197 2014
[19] J-M Morel and G Yu ldquoAsift A new framework for fullyaffine invariant image comparisonrdquo SIAM Journal on ImagingSciences vol 2 no 2 pp 438ndash469 2009
[20] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo in Proceedings of the IEEE Computer SocietyConference on Computer Vision and Pattern Recognition vol 2pp II-257ndashII-263 Madison WI USA June 2003
[21] B Zhang Y Jiao Z Ma Y Li and J Zhu ldquoAn efficientimage matching method using Speed Up Robust Featuresrdquoin Proceedings of the 11th IEEE International Conference onMechatronics and Automation IEEE ICMA 2014 pp 553ndash558China August 2014
[22] B Karasfi T S Hong A Jalalian and D Nakhaeinia ldquoSpeedupRobust Features based unsupervised place recognition forassistive mobile robotrdquo in Proceedings of the 2011 International
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
24 Computational and Mathematical Methods in Medicine
Conference on Pattern Analysis and Intelligent Robotics ICPAIR2011 pp 97ndash102 Malaysia June 2011
[23] H Bay A Ess T Tuytelaars and L Van Gool ldquoSpeeded-uprobust features (surf)rdquoComputer Vision and Image Understand-ing vol 110 no 3 pp 346ndash359 2008
[24] T Ojala M Pietikainen and T Maenpaa ldquoMultiresolutiongray-scale and rotation invariant texture classificationwith localbinary patternsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 24 no 7 pp 971ndash987 2002
[25] T Ojala M Pietikainen and T Maenpaa ldquoA generalized localbinary pattern operator for multiresolution gray scale androtation invariant texture classificationrdquo in Proceedings of theSecond International Conference on Advances in Pattern Recog-nition (ICAPR rsquo01) pp 397ndash406 Springer-Verlag London UK2001
[26] T Ahonen J Matas C He andM Pietikainen Rotation Invari-ant Image Description with Local Binary Pattern HistogramFourier Features pp 61ndash70 Springer Berlin Heidelberg BerlinHeidelberg Germany 2009
[27] G Zhao and M Pietikainen ldquoDynamic texture recognitionusing local binary patterns with an application to facial expres-sionsrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 29 no 6 pp 915ndash928 2007
[28] M Calonder V Lepetit M Ozuysal T Trzcinski C Strechaand P Fua ldquoBRIEF computing a local binary descriptorvery fastrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 34 no 7 pp 1281ndash1298 2012
[29] D Gong S Li and Y Xiang ldquoFace recognition using theWeberLocal Descriptorrdquo in Proceedings of the 1st Asian Conference onPattern Recognition ACPR 2011 pp 589ndash592 China November2011
[30] J Chen S Shan C He et al ldquoWLD A robust local imagedescriptorrdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 32 no 9 pp 1705ndash1720 2010
[31] S H Davarpanah F Khalid L Nurliyana Abdullah andM Golchin ldquoA texture descriptor BackGround Local BinaryPattern (BGLBP)rdquo Multimedia Tools and Applications vol 75no 11 pp 6549ndash6568 2016
[32] M Heikkila M Pietikainen and C Schmid Description ofInterest Regions with Center-Symmetric Local Binary Patternspp 58ndash69 Springer Berlin Heidelberg Berlin HeidelbergGermany 2006
[33] G Xue L Song J Sun and M Wu ldquoHybrid center-symmetriclocal pattern for dynamic background subtractionrdquo in Pro-ceedings of the 2011 12th IEEE International Conference onMultimedia and Expo (ICME rsquo11) pp 1ndash6 July 2011
[34] H Wu N Liu X Luo J Su and L Chen ldquoReal-timebackground subtraction-based video surveillance of people byintegrating local texture patternsrdquo Signal Image and VideoProcessing vol 8 no 4 pp 665ndash676 2014
[35] L Liu P Fieguth G Zhao M Pietikainen and D HuldquoExtended local binary patterns for face recognitionrdquo Informa-tion Sciences vol 358-359 pp 56ndash72 2016
[36] T Maenpaa and M Pietikainen ldquoClassification with color andtexture jointly or separatelyrdquo Pattern Recognition vol 37 no8 pp 1629ndash1640 2004
[37] G Xue J Sun and L Song ldquoDynamic background subtractionbased on spatial extended center-symmetric local binary pat-ternrdquo in Proceedings of the 2010 IEEE International ConferenceonMultimedia and Expo ICME 2010 pp 1050ndash1054 SingaporeJuly 2010
[38] S Liao G Zhao V Kellokumpu M Pietikainen and S Z LildquoModeling pixel process with scale invariant local patterns forbackground subtraction in complex scenesrdquo in Proceedings ofthe 2010 IEEE Computer Society Conference on Computer Visionand Pattern Recognition CVPR 2010 pp 1301ndash1306 USA June2010
[39] C Silva T Bouwmans and C Frelicot ldquoAn extended center-symmetric local binary pattern for background modeling andsubtraction in videosrdquo in Proceedings of the 10th InternationalConference on Computer Vision Theory and Applications (VIS-APP rsquo15) vol 1 pp 395ndash402 2015
[40] Y Chen L Ling andQ Huang ldquoClassification of breast tumorsin ultrasound using biclustering mining and neural networkrdquoin Proceedings of the 9th International Congress on Imageand Signal Processing BioMedical Engineering and InformaticsCISP-BMEI 2016 pp 1787ndash1791 China October 2016
[41] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning A review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006
[42] K T Rajakeerthana C Velayutham and K Thangavel Mam-mogram Image Classification Using Rough Neural Network pp133ndash138 Springer India New Delhi Indina 2014
[43] V Lessa and M Marengoni Applying Artificial Neural Networkfor the Classification of Breast Cancer Using Infrared Thermo-graphic Images pp 429ndash438 Springer International PublishingCham Germany 2016
[44] S Wan H-C Lee X Huang et al ldquoIntegrated local binarypattern texture features for classification of breast tissue imagedby optical coherence microscopyrdquo Medical Image Analysis vol38 pp 104ndash116 2017
[45] S M L de Lima A G da Silva-Filho and W P dos SantosldquoDetection and classification of masses in mammographicimages in a multi-kernel approachrdquo Computer Methods andPrograms in Biomedicine vol 134 pp 11ndash29 2016
[46] C Abirami R Harikumar and S Chakravarthy ldquoPerformanceanalysis and detection of micro calcification in digital mammo-grams usingwavelet featuresrdquo in Proceedings of the InternationalConference on Wireless Communications Signal Processing andNetworking (WiSPNET rsquo16) pp 2327ndash2331 Chennai IndiaMarch 2016
[47] N El Atlas A Bybi and H Drissi ldquoFeatures fusion forcharacterizing INBREAST-database massesrdquo in Proceedings ofthe 2nd International Conference on Electrical and InformationTechnologies ICEIT 2016 pp 374ndash379 Morocco May 2016
[48] H Alharbi G Falzon and P Kwan ldquoA novel feature reductionframework for digital mammogram image classificationrdquo inProceedings of the 3rd IAPR Asian Conference on PatternRecognition ACPR 2015 pp 221ndash225Malaysia November 2016
[49] W Peng R V Mayorga and E M A Hussein ldquoAn automatedconfirmatory system for analysis of mammogramsrdquo ComputerMethods and Programs in Biomedicine vol 125 pp 134ndash1442016
[50] A Jalalian S Mashohor R Mahmud B Karasfi M IqbalSaripan and A R Ramli ldquoComputer-assisted diagnosis systemfor breast cancer in computed tomography lasermammography(ctlm)rdquo Journal of Digital Imaging pp 1ndash16 2017
[51] H Li X Meng T Wang Y Tang and Y Yin ldquoBreast massesin mammography classification with local contour featuresrdquoBiomedical Engineering Online vol 16 no 1 44 pages 2017
[52] D-R Chen R-F Chang and Y-L Huang ldquoComputer-aideddiagnosis applied to US of solid breast nodules by using neuralnetworksrdquo Radiology vol 213 no 2 pp 407ndash412 1999
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Computational and Mathematical Methods in Medicine 25
[53] D-R Chen R-F Chang Y-L Huang Y-H Chou C-M Tiuand P-P Tsai ldquoTexture analysis of breast tumors on sonogramsrdquoSeminars in Ultrasound CT andMRI vol 21 no 4 pp 308ndash3162000
[54] N D Marom L Rokach and A Shmilovici ldquoUsing the confu-sion matrix for improving ensemble classifiersrdquo in Proceedingsof the 2010 IEEE 26th Convention of Electrical and ElectronicsEngineers in Israel IEEEI 2010 pp 555ndash559 Israel November2010
[55] S B Kotsiantis ldquoSupervised machine learning a review ofclassification techniquesrdquo in Proceedings of the 2007 Conferenceon Emerging Artificial Intelligence Applications in ComputerEngineering Real Word AI Systems with Applications in eHealthHCI Information Retrieval and Pervasive Technologies pp 3ndash242007
[56] F Rosenblatt The Perceptron A Perceiving and RecognizingAutomaton Cornell Aeronautical Laboratory Buffalo NewYork USA 1957
[57] K Hornik M Stinchcombe and HWhite ldquoMultilayer feedfor-ward networks are universal approximatorsrdquo Neural Networksvol 2 no 5 pp 359ndash366 1989
[58] R Hecht-Nielsen ldquoNeural networks for perceptionrdquo in chTheory of the Backpropagation Neural Network vol 2 pp 65ndash93 Harcourt Brace Co Orlando FL USA 1992
[59] J Li J H Cheng J Y Shi and F Huang ldquoBrief introductionof back propagation (BP) neural network algorithm and itsimprovementrdquo in Advances in Computer Science and Informa-tion EngineeringmdashVolume 2 D Jin and S Lin Eds vol 169of Advances in Intelligent and Soft Computing pp 553ndash558Springer Berlin Germany 2012
[60] A Dawson R Austin Jr and DWeinberg ldquoNuclear grading ofbreast carcinoma by image analysis Classification bymultivari-ate and neural network analysisrdquo American Journal of ClinicalPathology vol 95 Supplement 1 no 4 pp S29ndashS37 1991
[61] D-R Chen R-F Chang W-J Kuo M-C Chen and Y-LHuang ldquoDiagnosis of breast tumors with sonographic textureanalysis using wavelet transform and neural networksrdquo Ultra-sound inMedicine amp Biology vol 28 no 10 pp 1301ndash1310 2002
[62] S D De S Silva M G F Costa W C De A Pereira and CF F C Filho ldquoBreast tumor classification in ultrasound imagesusing neural networks with improved generalization methodsrdquoin Proceedings of the 37th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC rsquo15)pp 6321ndash6325 Italy August 2015
[63] I Saritas ldquoPrediction of breast cancer using artificial neuralnetworksrdquo Journal of Medical Systems vol 36 no 5 pp 2901ndash2907 2012
[64] E Lopez-Melendez L D Lara-Rodrıguez E Lopez-OlazagastiB Sanchez-Rinza and E Tepichin-Rodrıguez ldquoBICAD Breastimage computer aided diagnosis for standardBIRADS 1 and 2 incalcificationsrdquo in Proceedings of the 22nd Annual InternationalConference on Electronics Communications and ComputersCONIELECOMP 2012 pp 190ndash195 Mexico February 2012
[65] httpsgithubcomBVLCcaffe[66] Y Jia E Shelhamer J Donahue et al ldquoCaffe convolutional
architecture for fast feature embeddingrdquo CoRR 2014[67] httptorchch[68] httpwwwvlfeatorgmatconvnet[69] A Vedaldi and K Lenc ldquoMatconvnet - convolutional neural
networks for MATLABrdquo CoRR 2014[70] httpdeeplearningnetsoftwaretheano
[71] J Bergstra O Breuleux F Bastien et al ldquoTheano A cpu andgpu math compiler in pythonrdquo in Proceedings of the 9th Pythonin Science Conference pp 3ndash10 2010
[72] httpswwwtensorfloworg[73] httpsgithubcomMicrosoftCNTK[74] httpskerasio[75] httpsgithubcomml4j[76] httpceitautacirkeyvanradDeeBNet[77] M A Keyvanrad and M M Homayounpour ldquoA brief survey
on deep belief networks and introducing a new object orientedMATLAB toolbox (deebnet)rdquo CoRR vol abs14083264 2014
[78] C Y Wu S-C B Lo M T Freedman A Hasegawa R AZuurbier and S K Mun ldquoClassification of microcalcificationsin radiographs of pathological specimen for the diagnosis ofbreast cancerrdquo in Proceedings of the Medical Imaging pp 630ndash641 SPIE Digital Library Newport Beach CA USA 1994
[79] B Sahiner H-P Chan N Petrick et al ldquoClassification of massand normal breast tissue a convolution neural network classi-fier with spatial domain and texture imagesrdquo IEEE Transactionson Medical Imaging vol 15 no 5 pp 598ndash610 1996
[80] S-C B Lo H Li Y Wang L Kinnard and M T FreedmanldquoA multiple circular path convolution neural network systemfor detection of mammographic massesrdquo IEEE Transactions onMedical Imaging vol 21 no 2 pp 150ndash158 2002
[81] P Fonseca J Mendoza J Wainer et al ldquoAutomatic breastdensity classification using a convolutional neural networkarchitecture search procedurerdquo in Proceedings of the SPIEMedical Imaging Symposium 2015 Computer-Aided Diagnosisvol 9414 pp 941428ndash941428ndash8 USA February 2015
[82] J Arevalo F A Gonzalez R Ramos-Pollan J L Oliveiraand M A Guevara Lopez ldquoRepresentation learning for mam-mography mass lesion classification with convolutional neuralnetworksrdquo Computer Methods and Programs in Biomedicinevol 127 pp 248ndash257 2016
[83] H Su F Liu Y Xie F Xing S Meyyappan and L YangldquoRegion segmentation in histopathological breast cancer imagesusing deep convolutional neural networkrdquo in Proceedings of the12th IEEE International Symposium on Biomedical Imaging ISBI2015 pp 55ndash58 USA April 2015
[84] K Sharma and B Preet ldquoClassification of mammogram imagesby using CNN classifierrdquo in Proceedings of the 5th InternationalConference on Advances in Computing Communications andInformatics ICACCI 2016 pp 2743ndash2749 India September2016
[85] H Rezaeilouyeh A Mollahosseini andM HMahoor ldquoMicro-scopic medical image classification framework via deep learn-ing and shearlet transformrdquo Journal of Medical Imaging vol 3no 4 Article ID 044501 2016
[86] A Albayrak and G Bilgin Mitosis Detection Using Convolu-tional Neural Network Based Features pp 335ndash340 2017
[87] Z Jiao X Gao Y Wang and J Li ldquoA deep feature basedframework for breast masses classificationrdquo Neurocomputingvol 197 pp 221ndash231 2016
[88] M Zejmo M Kowal J Korbicz and R Monczak ldquoClassifica-tion of breast cancer cytological specimen using convolutionalneural networkrdquo Journal of Physics Conference Series vol 783no 1 Article ID 012060 2017
[89] F Jiang H Liu S Yu and Y Xie ldquoBreast mass lesion classifi-cation in mammograms by transfer learningrdquo in Proceedings ofthe 5th International Conference on Bioinformatics and Compu-tational Biology (ICBCB rsquo17) pp 59ndash62 ACM New York NYUSA 2017
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
26 Computational and Mathematical Methods in Medicine
[90] S Suzuki X Zhang N Homma et al ldquoMass detectionusing deep convolutional neural network for mammographiccomputer-aided diagnosisrdquo in Proceedings of the 55th AnnualConference of the Society of Instrument and Control Engineersof Japan (SICE rsquo16) pp 1382ndash1386 Japan September 2016
[91] Y Qiu Y Wang S Yan et al ldquoAn initial investigation ondeveloping a new method to predict short-term breast cancerrisk based on deep learning technologyrdquo in Proceedings of theMedical Imaging 2016 Computer-Aided Diagnosis SPIE DigitalLibrary San Diego California USA March 2016
[92] R K Samala H-P Chan L M Hadjiiski K Cha andM A Helvie ldquoDeep-learning convolution neural networkfor computer-aided detection of microcalcifications in digitalbreast tomosynthesisrdquo in Proceedings of the Medical Imaging2016 Computer-Aided Diagnosis USA March 2016
[93] T Kooi G Litjens B van Ginneken et al ldquoLarge scaledeep learning for computer aided detection of mammographiclesionsrdquoMedical Image Analysis vol 35 pp 303ndash312 2017
[94] K J Geras S Wolfson S G Kim L Moy and K Cho ldquoHigh-resolution breast cancer screening withmulti-view deep convo-lutional neural networksrdquo CoRR vol abs170307047 2017
[95] S Beura B Majhi R Dash and S Roy ldquoClassification ofmammogram using two-dimensional discrete orthonormal S-transform for breast cancer detectionrdquo Healthcare TechnologyLetters vol 2 no 2 pp 46ndash51 2015
[96] J Diz G Marreiros and A Freitas Using Data MiningTechniques to Support Breast Cancer Diagnosis Advances inIntelligent Systems and Computing pp 689ndash700 SpringerInternational Publishing Cham Switzerland 2015
[97] J Zhang J I Silber and M A Mazurowski ldquoModelingfalse positive error making patterns in radiology trainees forimproved mammography educationrdquo Journal of BiomedicalInformatics vol 54 pp 50ndash57 2015
[98] F K Ahmad and N Yusoff ldquoClassifying breast cancer typesbased on fine needle aspiration biopsy data using random forestclassifierrdquo in Proceedings of the 2013 13th International Confer-ence on Intellient SystemsDesign andApplications (ISDA rsquo13) pp121ndash125 Malaysia December 2013
[99] A Paul A Dey D P Mukherjee J Sivaswamy and V TouraniRegenerative Random Forest with Automatic Feature Selectionto Detect Mitosis in Histopathological Breast Cancer Images vol9350 of Lecture Notes in Computer Science pp 94ndash102 SpringerInternational Publishing Cham Switzerland 2015
[100] Z Chen M Berks S Astley and C Taylor Classification ofLinear Structures in Mammograms Using Random Forests Lec-ture Notes in Computer Science pp 153ndash160 Springer BerlinHeidelberg Heidelberg Germany 2010
[101] Y Zhang B Zhang and W Lu ldquoBreast cancer classificationfrom histological images with multiple features and randomsubspace classifier ensemblerdquo in Proceedings of the 2011 Inter-national Symposium on Computational Models for Life Sciences(CMLS rsquo11) vol 1371 of AIP Conference Proceedings pp 19ndash282011
[102] S P Angayarkanni and N B Kamal ldquoMRI mammogram imageclassification using ID3 algorithmrdquo in Proceedings of the IETConference on Image Processing (IPR rsquo12) pp 1ndash5 IET LondonUK July 2012
[103] K Wang M Dong Z Yang Y Guo and Y Ma ldquoRegions ofmicro-calcifications clusters detection based on new featuresfrom imbalance data in mammogramsrdquo in Proceedings of the
2016 8th International Conference on Graphic and Image Pro-cessing (ICGIP rsquo16) vol 10225 pp 102252Cndash102252Cndash6 SPIEDigital Library Tokyo Japan 2017
[104] D O Tambasco Bruno M Z Do Nascimento R P Ramos VR Batista L A Neves and A S Martins ldquoLBP operators oncurvelet coefficients as an algorithm to describe texture in breastcancer tissuesrdquo Expert Systems with Applications vol 55 pp329ndash340 2016
[105] C Muramatsu T Hara T Endo and H Fujita ldquoBreast massclassification on mammograms using radial local ternary pat-ternsrdquo Computers in Biology and Medicine vol 72 pp 43ndash532016
[106] MDong X Lu YMa Y Guo YMa andKWang ldquoAn efficientapproach for automated mass segmentation and classificationin mammogramsrdquo Journal of Digital Imaging vol 28 no 5 pp613ndash625 2015
[107] G Piantadosi R Fusco A PetrilloM Sansone andC SansoneLBP-TOP for Volume Lesion Classification in Breast DCE-MRI pp 647ndash657 Springer International Publishing ChamSwitzerland 2015
[108] B Malik J Klock J Wiskin and M Lenox ldquoObjective breasttissue image classification using Quantitative Transmissionultrasound tomographyrdquo Scientific Reports vol 6 no 3 ArticleID 38857 2016
[109] R-F Chang W-J Wu W K Moon Y-H Chou and D-RChen ldquoSupport vector machines for diagnosis of breast tumorson US imagesrdquo Academic Radiology vol 10 no 2 pp 189ndash1972003
[110] C Akbay N G Gencer and G Gencer ldquoCAD for detectionof microcalcification and classification in Mammogramsrdquo inProceedings of the 2014 18th National Biomedical EngineeringMeeting (BIYOMUT rsquo14) pp 1ndash4 Turkey October 2014
[111] J Levman T Leung P Causer D Plewes and A L Mar-tel ldquoClassification of dynamic contrast-enhanced magneticresonance breast lesions by support vector machinesrdquo IEEETransactions on Medical Imaging vol 27 no 5 pp 688ndash6962008
[112] L de Oliveira Martins E C da Silva A C Silva A C de Paivaand M Gattass ldquoClassification of Breast Masses in Mammo-gram Images Using Ripleyrsquos K Function and Support VectorMachinerdquo in Machine Learning and Data Mining in PatternRecognition vol 4571 of Lecture Notes in Computer Sciencepp 784ndash794 Springer Berlin Heidelberg Berlin HeidelbergGermany 2007
[113] K Fukushima ldquoNeocognitron a self-organizing neural net-work model for a mechanism of pattern recognition unaffectedby shift in positionrdquo Biological Cybernetics vol 36 no 4 pp193ndash202 1980
[114] A Krizhevsky I Sutskever and G E Hinton ldquoImagenet classi-fication with deep convolutional neural networksrdquo in Advancesin Neural Information Processing Systems 25 F Pereira C J CBurges L Bottou and K Q Weinberger Eds pp 1097ndash1105Curran Associates Inc 2012
[115] C Szegedy W Liu Y Jia et al ldquoGoing deeper with convolu-tionsrdquo CoRR vol abs14094842 2014
[116] K He X Zhang S Ren and J Sun ldquoDeep residual learning forimage recognitionrdquo CoRR vol abs151203385 2015
[117] C Szegedy V Vanhoucke S Ioffe J Shlens and Z WojnaldquoRethinking the inception architecture for computer visionrdquoCoRR vol abs151200567 2015
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Computational and Mathematical Methods in Medicine 27
[118] C Szegedy S Ioffe and V Vanhoucke ldquoInception-v4inception-resnet and the impact of residual connections onlearningrdquo CoRR vol abs160207261 2016
[119] N Tajbakhsh J Y Shin S R Gurudu et al ldquoConvolutionalneural networks for medical image analysis full training or finetuningrdquo IEEE Transactions on Medical Imaging vol 35 no 5pp 1299ndash1312 2016
[120] Y Liu S Zhou and Q Chen ldquoDiscriminative deep beliefnetworks for visual data classificationrdquo Pattern Recognition vol44 no 10-11 pp 2287ndash2296 2011
[121] A M Abdel-Zaher and A M Eldeib ldquoBreast cancer clas-sification using deep belief networksrdquo Expert Systems withApplications vol 46 pp 139ndash144 2016
[122] Y-D Zhang S-H Wang G Liu and J Yang ldquoComputer-aided diagnosis of abnormal breasts in mammogram imagesby weighted-type fractional Fourier transformrdquo Advances inMechanical Engineering vol 8 no 2 pp 1ndash11 2016
[123] F Shirazi and E Rashedi ldquoDetection of cancer tumors inmammography images using support vector machine andmixed gravitational search algorithmrdquo in Proceedings of the 1stConference on Swarm Intelligence and Evolutionary Computa-tion (CSIEC rsquo16) pp 98ndash101 Iran March 2016
[124] M Sewak P Vaidya C-C Chan and Z-H Duan ldquoSVMapproach to breast cancer classificationrdquo in Proceedings of the2nd International Multi-Symposiums on Computer and Compu-tational Sciences 2007 (IMSCCS rsquo07) pp 32ndash37 IEEE Iowa CityIA USA August 2007
[125] J Dheeba and S Tamil Selvi ldquoClassification of malignant andbenign microcalcification using SVM classifierrdquo in Proceedingsof the 2011 International Conference on Emerging Trends in Elec-trical and Computer Technology (ICETECT rsquo11) pp 686ndash690India March 2011
[126] M Taheri GHamer S H Son and S Y Shin ldquoEnhanced breastcancer classification with automatic thresholding using SVMand Harris corner detectionrdquo in Proceedings of the InternationalConference on Research in Adaptive and Convergent Systems(RACS rsquo16) pp 56ndash60 ACM Odense Denmark October 2016
[127] M Tan J Pu and B Zheng ldquoOptimization of breast mass clas-sification using sequential forward floating selection (SFFS) anda support vector machine (SVM) modelrdquo International Journalfor Computer Assisted Radiology and Surgery vol 9 no 6 pp1005ndash1020 2014
[128] S Kavitha and K K Thyagharajan ldquoFeatures based mam-mogram image classification using weighted feature supportvectormachinerdquoCommunications in Computer and InformationScience vol 270 no II pp 320ndash329 2012
[129] E J Kendall and M T Flynn ldquoAutomated breast imageclassification using features from its discrete cosine transformrdquoPLoS ONE vol 9 no 3 Article ID e91015 pp 1ndash8 2014
[130] V Oleksyuk F Saleheen D F Caroline S A Pascarella and C-H Won ldquoClassification of breast masses using Tactile ImagingSystem and machine learning algorithmsrdquo in Proceedings of the2016 IEEE Signal Processing inMedicine and Biology Symposium(SPMB rsquo16) pp 1ndash4 USA Dec 2016
[131] F Burling-Claridge M Iqbal and M Zhang ldquoEvolutionaryalgorithms for classification of mammographie densities usinglocal binary patterns and statistical featuresrdquo in Proceedings ofthe 2016 IEEE Congress on Evolutionary Computation (CEC rsquo16)pp 3847ndash3854 Canada July 2016
[132] U Raghavendra U Rajendra Acharya H Fujita A Gudigar JH Tan and S Chokkadi ldquoApplication of Gabor wavelet and
Locality Sensitive Discriminant Analysis for automated identi-fication of breast cancer using digitized mammogram imagesrdquoApplied Soft Computing vol 46 pp 151ndash161 2016
[133] N P Perez M A Guevara Lopez A Silva and I RamosldquoImproving the Mann-Whitney statistical test for feature selec-tion an approach in breast cancer diagnosis onmammographyrdquoArtificial Intelligence in Medicine vol 63 no 1 pp 19ndash31 2015
[134] G D Rashmi A Lekha and N Bawane ldquoAnalysis of efficiencyof classification and prediction algorithms (Naıve Bayes) forBreast Cancer datasetrdquo in Proceedings of the 2015 InternationalConference on Emerging Research in Electronics Computer Sci-ence and Technology (ICERECT rsquo15) pp 108ndash113 IEEEMandyaIndia December 2015
[135] G Gatuha and T Jiang ldquoAndroid based Naive Bayes proba-bilistic detection model for breast cancer and Mobile CloudComputing Design and Implementationrdquo International Journalof Engineering Research in Africa vol 21 pp 197ndash208 2016
[136] M Benndorf E Kotter M Langer C Herda Y Wu and E SBurnside ldquoDevelopment of an online publicly accessible naiveBayesian decision support tool formammographicmass lesionsbased on the American College of Radiology (ACR) BI-RADSlexiconrdquo European Radiology vol 25 no 6 pp 1768ndash1775 2015
[137] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoImproving bayesiannetworks breast mass diagnosis by using clinical datardquo LectureNotes in Computer Science (including subseries Lecture Notes inArtificial Intelligence and Lecture Notes in Bioinformatics) vol9116 pp 292ndash301 2015
[138] K A Nugroho N A Setiawan and T B Adji ldquoCascade gener-alization for breast cancer detectionrdquo in Proceedings of the 20135th International Conference on Information Technology andElectrical Engineering (ICITEE rsquo13) pp 57ndash61 IEEE YogyakartaIndonesia October 2013
[139] V Rodrıguez-Lopez and R Cruz-Barbosa ldquoOn the breast massdiagnosis using Bayesian networksrdquo Lecture Notes in ComputerScience (including subseries LectureNotes inArtificial Intelligenceand Lecture Notes in Bioinformatics) vol 8857 pp 474ndash4852014
[140] S Sivakumari R Praveena Priyadarsini and P AmudhaldquoAccuracy evaluation of C45 and Naıve Bayes classifiers usingattribute ranking methodrdquo International Journal of Computa-tional Intelligence Systems vol 2 no 1 pp 60ndash68 2009
[141] V Rodrıguez-Lopez and R Cruz-Barbosa Improving BayesianNetworks Breast Mass Diagnosis by Using Clinical Data pp292ndash301 Springer International Publishing Cham Switzerland2015
[142] H M Moftah A T Azar E T Al-Shammari N I Ghali A EHassanien andM Shoman ldquoAdaptive k-means clustering algo-rithm for MR breast image segmentationrdquo Neural Computingand Applications vol 24 no 7-8 pp 1917ndash1928 2014
[143] S H Lee J H Kim K G Kim S J Park and W K MoonK-Means Clustering and Classification of Kinetic Curves onMalignancy in Dynamic Breast MRI pp 2536ndash2539 SpringerBerlin Heidelberg Berlin Heidelberg Germany 2007
[144] S Dalmiya A Dasgupta and S Kanti Datta ldquoApplication ofWavelet based K-means Algorithm in Mammogram Segmen-tationrdquo International Journal of Computer Applications vol 52no 15 pp 15ndash19 2012
[145] A Elmoufidi K El Fahssi S J Andaloussi and A SekkakildquoDetection of regions of interest inmammograms by using localbinary pattern and dynamicK-means algorithmrdquoOrbAcademicPublisher 2014
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
28 Computational and Mathematical Methods in Medicine
[146] E S Samundeeswari P K Saranya and R Manavalan ldquoSeg-mentation of Breast Ultrasound image using Regularized K-Means (ReKM) clusteringrdquo in Proceedings of the 2016 IEEEInternational Conference on Wireless Communications SignalProcessing and Networking (WiSPNET rsquo16) pp 1379ndash1383 IndiaMarch 2016
[147] J H K Rezaee ldquoDesigning an algorithm for cancerous tissuesegmentation using adaptive k-means cluttering and discretewavelet transformrdquo Journal of Biomedical Physics and Engineer-ing pp 93ndash104 2013
[148] B Chandra S Nath and A Malhothra ldquoClassification andclustering of breast cancer imagesrdquo in Proceedings of the Inter-national Joint Conference on Neural Networks 2006 (IJCNN rsquo06)pp 3843ndash3847 2006
[149] J R Quinlan ldquoInduction of decision treesrdquo Machine Learningvol 1 no 1 pp 81ndash106 1986
[150] J R Quinlan C45 Programs for Machine Learning MorganKaufmann Publishers Inc San Francisco CA USA 1993
[151] A I Pritom M A R Munshi S A Sabab and S ShihabldquoPredicting breast cancer recurrence using effective classifica-tion and feature selection techniquerdquo in Proceedings of the 19thInternational Conference on Computer and Information Technol-ogy (ICCIT rsquo16) pp 310ndash314 December 2016
[152] H Asri H Mousannif H Al Moatassime and T Noel ldquoUsingmachine learning algorithms for breast cancer risk predictionand diagnosisrdquo Procedia Computer Science vol 83 pp 1064ndash1069 2016
[153] L Breiman ldquoArcing classifiersrdquoThe Annals of Statistics vol 26no 3 pp 801ndash849 1998
[154] J H Friedman ldquoStochastic gradient boostingrdquo ComputationalStatistics and Data Analysis vol 38 Nonlinear Methods andData Mining no 4 pp 367ndash378 2002
[155] T Chen and C Guestrin ldquoXgboost a scalable tree boostingsystemrdquo CoRR vol abs160302754 2016
[156] I El-Naqa Y Yang M N Wernick N P Galatsanos and R MNishikawa ldquoA support vector machine approach for detectionof microcalcificationsrdquo IEEE Transactions on Medical Imagingvol 21 no 12 pp 1552ndash1563 2002
[157] R-F ChangW-J WuW KMoon and D-R Chen ldquoImprove-ment in breast tumor discrimination by support vectormachines and speckle-emphasis texture analysisrdquoUltrasound inMedicine amp Biology vol 29 no 5 pp 679ndash686 2003
[158] Y Chu L Li D Goldgof Y Qiu and R A Clark ldquoClassificationof masses on mammograms using support vector machinerdquo inProceedings of the Medical Imaging 2003 Image Processing pp940ndash948 USA February 2003
[159] B K Singh K Verma A Thoke and J S Suri ldquoRisk stratifica-tion of 2D ultrasound-based breast lesions using hybrid featureselection inmachine learning paradigmrdquoMeasurement vol 105pp 146ndash157 2017
[160] A Lashkari andM Firouzmand ldquoEarly breast cancer detectionin thermogram images using AdaBoost classifier and fuzzy C-Means clustering algorithmrdquoMiddle East Journal of Cancer vol7 no 3 pp 113ndash124 2016
[161] T W Nattkemper B Arnrich O Lichte et al ldquoEvaluation ofradiological features for breast tumour classification in clinicalscreening with machine learning methodsrdquo Artificial Intelli-gence in Medicine vol 34 no 2 pp 129ndash139 2005
[162] L A Salazar-Licea J C Pedraza-Ortega A Pastrana-PalmaandMA Aceves-Fernandez ldquoLocation ofmammogramsROIrsquosand reduction of false-positiverdquo Computer Methods and Pro-grams in Biomedicine vol 143 pp 97ndash111 2017
[163] K D Marcomini A A O Carneiro and H Schiabel ldquoAppli-cation of artificial neural network models in segmentation andclassification of nodules in breast ultrasound digital imagesrdquoInternational Journal of Biomedical Imaging vol 2016 ArticleID 7987212 13 pages 2016
[164] D-R Chen R-F Chang and Y-L Huang ldquoBreast cancer diag-nosis using self-organizing map for sonographyrdquo Ultrasound inMedicine amp Biology vol 26 no 3 pp 405ndash411 2000
[165] Z Iscan Z Dokur and T Olmez Improved Incremental Self-Organizing Map forThe Segmentation of Ultrasound Images pp293ndash302 Springer Netherlands Dordrecht Netherlands 2007
[166] F R Cordeiro W P Santos and A G Silva-Filho ldquoA semi-supervised fuzzy GrowCut algorithm to segment and classifyregions of interest of mammographic imagesrdquo Expert Systemswith Applications vol 65 pp 116ndash126 2016
[167] F R Cordeiro W P Santos and A G Silva-Filho ldquoAnal-ysis of supervised and semi-supervised GrowCut applied tosegmentation of masses in mammography imagesrdquo ComputerMethods in Biomechanics and Biomedical Engineering Imagingand Visualization vol 5 no 4 pp 297ndash315 2017
[168] Z Nawel A Nabiha D Nilanjan and S Mokhtar ldquoAdaptivesemi supervised support vectormachine semi supervised learn-ing with features cooperation for breast cancer classificationrdquoJournal of Medical Imaging and Health Informatics vol 6 no 1pp 53ndash62 2016
[169] N Zemmal N Azizi and M Sellami ldquoCAD system forclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithm and heterogeneousfeaturesrdquo in Proceedings of the 12th International Symposium onProgramming and Systems (ISPS rsquo15) pp 245ndash253 IEEEAlgiersAlgeria April 2015
[170] N ZemmalNAzizi NDey andM Sellami ldquoAdaptative S3VMsemi supervised learning with features cooperation for breastcancer classificationrdquo Journal of Medical Imaging and HealthInformatics vol 6 no 4 pp 957ndash967 2016
[171] N Zemmal N Azizi M Sellami and N Dey ldquoAutomatedclassification of mammographic abnormalities using transduc-tive semi supervised learning algorithmrdquo in Proceedings of theMediterranean Conference on Information amp CommunicationTechnologies 2015 A El Oualkadi F Choubani and A ElMoussati Eds pp 657ndash662 Springer International PublishingCham 2016
[172] M Peikari J Zubovits G Clarke and A L Martel ldquoClusteringanalysis for semi-supervised learning improves classificationperformance of digital pathologyrdquo in Proceedings of the Inter-national Workshop on Machine Learning in Medical ImagingMICCAI 2015 vol 9352 of Lecture Notes in Computer Sciencepp 263ndash270 Springer International Publishing Cham Switzer-land 2015
[173] MAnkerstMMBreunigH-PKriegel and J Sander ldquoOpticsOrdering points to identify the clustering structurerdquo SIGMODRec vol 28 pp 49ndash60 June 1999
[174] Y Zhu F Li T J Vadakkan et al ldquoThree-dimensional vas-culature reconstruction of tumour microenvironment via localclustering and classificationrdquo Interface Focus vol 3 no 4 2013
[175] X Liu J Shi S Zhou and M Lu ldquoAn iterated Laplacian basedsemi-supervised dimensionality reduction for classification ofbreast cancer on ultrasound imagesrdquo in Proceedings of the 201436th Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBC rsquo14) pp 4679ndash4682 USAAugust 2014
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016
Computational and Mathematical Methods in Medicine 29
[176] J Ding H D Cheng J Huang J Liu and Y Zhang ldquoBreastultrasound image classification based on multiple-instancelearningrdquo Journal of Digital Imaging vol 25 no 5 pp 620ndash6272012
[177] M Pang Y Wang and J Li ldquoDirichlet-based concentric circlefeature transform for breast mass classificationrdquo in Proceedingsof the 27th IEEE International Conference on Tools with ArtificialIntelligence (ICTAI 2015) vol 2016 pp 272ndash277 IEEE Vietri sulMare Italy November 2015
[178] N C Mhala and S H Bhandari ldquoImproved approach towardsclassification of histopathology images using bag-of-featuresrdquoin Proceedings of the 2016 International Conference on Signal andInformation Processing (IConSIP rsquo16) IEEE Vishnupuri IndiaOctober 2016
[179] C Hiba Z Hamid and A Omar ldquoAn improved breast tissuedensity classification framework using bag of features modelrdquoin Proceedings of the 4th IEEE International Colloquium onInformation Science and Technology CiSt 2016 pp 405ndash409Morocco October 2016
[180] P Langley W Iba and K Thompson ldquoAnalysis of Bayesianclassifiersrdquo in Proceedings of the 10th National Conference onArtificial Intelligence (AAAI rsquo92) pp 223ndash228 AAAI Press SanJose Calif USA July 1992
[181] A Tosun A B Bener and S Akbarinasaji ldquoA systematicliterature review on the applications of Bayesian networks topredict software qualityrdquo Software Quality Journal vol 25 no1 pp 273ndash305 2017
[182] J Grover ldquoA Literature Review of Bayesrsquo Theorem and BayesianBeliefNetworks (BBN)rdquo in Strategic EconomicDecision-Makingvol 9 of SpringerBriefs in Statistics pp 11ndash27 SpringerNewYorkNew York NY 2013
[183] S M Butler G I Webb and R A Lewis ldquoA case study infeature invention for breast cancer diagnosis using X-ray scatterimagesrdquo in AI 2003 advances in artificial intelligence vol 2903of Lecture Notes in Computer Science pp 677ndash685 SpringerBerlin Germany 2003
[184] E A Fischer J Y Lo and M K Markey ldquoBayesian networks ofBI-RADSspl trade descriptors for breast lesion classificationrdquoin Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (IEMBS rsquo04)vol 26 IV pp 3031ndash3034 IEEE San Francisco CA USA Sept2004
[185] D Soria J M Garibaldi E Biganzoli and I O Ellis ldquoA com-parison of three different methods for classification of breastcancer datardquo in Proceedings of the 7th International Conferenceon Machine Learning and Applications (ICMLA rsquo08) pp 619ndash624 USA December 2008
[186] httpwwwcswaikatoacnzmlweka[187] T Masquelier and S J Thorpe ldquoUnsupervised learning of
visual features through spike timing dependent plasticityrdquo PLoSComputational Biology vol 3 no 2 pp 1ndash11 2007
[188] J MacQueen ldquoSome methods for classification and analysis ofmultivariate observationsrdquo in Proceedings of the 5th BerkeleySymposium on Mathematical Statistics and Probability vol 1pp 281ndash297 University of California Press Berkeley Calif USA1967
[189] T Kohonen M R Schroeder and T S Huang Eds Self-Organizing Maps Springer-Verlag New York Secaucus NJUSA 3rd edition 2001
[190] T Kohonen ldquoEssentials of the self-organizing maprdquo NeuralNetworks vol 37 Twenty-fifth Anniversay CommemorativeIssue pp 52ndash65 2013
[191] T Kohonen ldquoThe Self-Organizing Maprdquo Proceedings of theIEEE vol 78 no 9 pp 1464ndash1480 1990
[192] J C Dunn ldquoA fuzzy relative of the ISODATA process and itsuse in detecting compact well-separated clustersrdquo Journal ofCybernetics vol 3 no 3 pp 32ndash57 1973
[193] J C Bezdek Pattern Recognition with Fuzzy Objective FunctionAlgorithms Kluwer Academic Publishers Norwell MA USA1981
[194] T C Cahoon M A Sutton and J C Bezdek ldquoBreast cancerdetection using image processing techniquesrdquo in Proceedingsof the FUZZ-IEEE 2000 9th IEEE International Conference onFuzzy Systems pp 973ndash976 May 2000
[195] M K Markey J Y Lo G D Tourassi and C E Floyd Jr ldquoSelf-organizing map for cluster analysis of a breast cancer databaserdquoArtificial Intelligence inMedicine vol 27 no 2 pp 113ndash127 2003
[196] X Zhu ldquoSemi-supervised learning literature surveyrdquo TechRep University of Wisconsin-Madison 2005 Tech Rep 1530Computer Sciences
[197] C H Li and P C Yuen ldquoSemi-supervised Learning in MedicalImage Databaserdquo inAdvances in Knowledge Discovery and DataMining vol 2035 of Lecture Notes in Computer Science pp 154ndash160 Springer Berlin Heidelberg Berlin Heidelberg Germany2001
[198] J-B Li Y Yu Z-M Yang and L-L Tang ldquoBreast tissue imageclassification based on semi-supervised locality discriminantprojection with kernelsrdquo Journal of Medical Systems vol 36 no5 pp 2779ndash2786 2012
[199] M Ngadi A Amine and B Nassih ldquoA robust approach formammographic image classification using NSVC algorithmrdquoin Proceedings of the 1st Mediterranean Conference on PatternRecognition and Artificial Intelligence (MedPRAI rsquo16) vol PartF126741 pp 44ndash49 Algeria November 2016