Brain Tumor Segmentation Using A Novel Unified Legendra ...joics.org/gallery/ics-1164.pdf · The Brain Tumor segmentation is a process of identifying affected tumor tissues and protects

Brain Tumor Segmentation Using A Novel Unified

Legendra Polynomial Algorithm (ULPA) in MRI

Images

C. Jaspin Jeba Sheela

Reg.No. 17221282162010,Research Scholar, St.Xavier’s Autonomous College,

Palayamkottai affiliated to Manonmaniam Sundaranar University, Abishekapatti,

Tirunelveli 627012, Tamil Nadu, India

E-mail: [email protected]

G. Suganthi

Associate Professor, Department of Computer Science, Women’s Christian College,

Nagercoil affiliated to Manonmaniam Sundaranar University, Abishekapatti, Tirunelveli

627012, Tamil Nadu, India

E-mail: [email protected]

Abstract

The segmentation, detection, and extraction of infected tumor area from

Magnetic Resonance MR Images are a prime concern as they are tedious and time

taking task performed by radiologists or clinical experts. This study analyzes the ways to

improve performance and reduce the complexity involved in the medical image

segmentation process. This paper describes a novel Unified Legendre Polynomial

Algorithm which is an important segmentation performance for automatic tumor

segmentation. In this paper, Spatial Fuzzy C-Means clustering is used to evaluate the

Region Of Interest (ROI) in MRI images. A two step approach is designed to upgrade the

tumor border with region merging and improved distance regularization level. BRATs

2015 training database, evaluates the accuracy and robustness of this method with

respect performance scores, Dice, Positive Predictive Value (PPV), Sensitivity,

Hausdorff Distance (HD) and Euclidean Distance (ED). In general, the proposed

method is effective in segmenting tumor in MRI images, and it has the potential to

identify the tumors in daily clinical, routine examination.

Keywords: MR Images, region growing, Legendre polynomial, Segmentation, Tumor

detection

Journal of Information and Computational Science

Volume 9 Issue 8 - 2019

ISSN: 1548-7741

www.joics.org270

mailto:[email protected]

1. Introduction

The brain is a highly specialized organ. It serves as the chief control mechanism

of the body. The brain is a soft, spongy mass of tissues. The brain is in charge of our

sensory organs. The brain has a very complex structure. The skull keeps the brain safe.

Tumors can directly destroy all healthy brain cells. The tumor detects the sudden growth

of a particular period.[12]

A Brain tumor is a formation of abnormal cells within the brain that can disrupt

the function of the brain. A Brain tumor is an uncontrolled growth of cells. When our

body functions in a normal manner, the cells die and get replaced by new cells. But this

normal cycle gets disrupted in tumors. Brain Tumor has various sizes, shapes, locations,

and appear in different image intensities. Manual detection of the tumor is a time

consuming task and is also inaccurate. There are many automated methods which can be

used for surgical and treatment planning. But they have specific drawbacks and

limitations.

Magnetic Resonance Imaging (MRI) is a type of scan that uses magnetic fields

and radio waves. MRI is the most common type of tests used to diagnose brain tumors.

It uses computers to create detailed images of the brain. The MRI is the best type of

brain tumor diagnosis than the others. It detects the brain tumors with high resolution and

ability to show clear brain structures [11]. MRI provides Brain and nerve tissues in

multiple planes without overlying bones. Medical images pose very important and useful

information about the anatomical structure of the human body.MRI is used for

visualizing the internal structure of the body.MRI provides rich information for brain

tumor diagnosis and treatment planning. MRI images also increase the difficulty in the

segmentation of tumor [1]. MRI is a challenging and critical task in Medical Image

Analysis. The advantage of MRI is that it has no radiation. MRI is a non invasive

medical image technique that provides high resolution images for the Structure [6]. MRI

images have been used frequently by radiologists. This study addresses the problems of

segmentation of abnormal brain tissues and normal tissues such as Gray Matter (GM),

White Matter (WM), seed point and Normalization from Magnetic Resonance (MR)

Images using feature extraction.

2. Literature Survey

The author Maddalena Strumia et al. [1] proposes the spatial lesion distribution

which plays a major role in diagnosing tumor segmentation based on an adaptive

geometric brain model. This is to motivate and formulate a new distance to evaluate the

quality of the brain tumor segmentation which shows the region of abnormalities. The

topological properties of the lesions and brain tissues are segmented as the white matter.

Paper [2] presents an automatic segmentation method based on Convolutional

Neural Networks (CNN) which explores the small 3X3 kernels. The small 3X3 Kernels

are used to design a deeper architecture and identify the use of intensity normalization

as a preprocessing step. It can consume much time to calculate the manual

segmentation.



ISSN: 1548-7741

www.joics.org271

Paper [3] suggests the accuracy and feature extraction of MRI brain tumor

segmentation. The advantages of this paper is a correlation between intracranial

structure deformation and compression from MRI brain Tumor growth. The techniques

used in this paper are 3-Dimensional non rigid registration and deformation modeling

techniques.

Bjoern H. Menze et al [4] represents a generative probabilistic model for

segmentation of brain lesions in multidimensional MR Images. Gaussian Mixture and

probabilistic tissue atlas methods estimate the label map for a new image. These

methods extract a latent atlas prior distribution and the lesion posterior distribution

jointly from the image data sets.

Annemie Ribbens et al [5] evaluates a large amount of MRI segmentation and

comparison followed by the normal and abnormal MRIs. The main advantage of this

paper is to identify homogeneous subgroups automatically in the unsupervised method

and detect the relevant morphological features based on the segmentation. The atlas

method is optimally adapted for guiding the segmentation of each subgroup.

Colm Elliott [6] represents two stages of classification process one is Bayesian

classifier which provides a probabilistic brain tissue and another one is a random forest

based lesion level classification and it compares the truth segmentation and the manual

identification of MRI.

Zexuan Ji [7] represents the voxel’s neighborhood which satisfies the Gaussian

Mixture Model (GMM) and fuzzy local GMM (FLGMM) algorithm for automated

brain MR Image segmentation. It compares to the algorithm to state of the art

segmentation approaches in both synthetic and clinical data. It overcomes the

difficulties raised by noise, low contrast, and bias field and improves the accuracy of

brain MR Images segmentations.

Y Chen [8] proposes a new energy minimization framework for simultaneous

estimation of the intensity inhomogeneities and segmentation. It is formulated to

modify the objective function of the standard fuzzy C Means algorithm and the

functions which depends on coefficient of the basis function, membership ratio,

centroid and non-local information of MR Images.

A.Ortiz [9] represents the diagnosis of brain disorders. The main disadvantage of

this paper is that it has discovered different regions on the image without using prior

information. It consists of hybridizing multi objective optimization for feature selection

with a Growing Hierarchical Self Organizing Map (GHSOM) classifier and a

probability clustering method.

The above literature survey has revealed that some of the techniques

obtain only segmentation while some of the other techniques are invented to obtain

identification, detection and Feature Extraction.

3. Proposed MRI Segmented Method

Brain tumor segmentation is an important and challenging factor in the medical

image segmentation. The Tumor is mainly categorized into two groups. They are

probability based methods and non-probability based method [4]. The probability based



ISSN: 1548-7741

www.joics.org272

methods directly learn the relationship between specific image features and

segmentation. This method is also used to find out the pixel labels and intensities. Non-

probability based segmentation method works efficiently. For example, Fuzzy C-

Means (FCM) algorithm is presented for automatic Tumor segmentation, to develop a

Legendre Polynomial Algorithm and to detect the Tumor borders in MRI brain Images.

It comprises of three components,

1) Estimate Region Of Interest (ROI) with fewer pseudo lesions.[1]

2) Detect the entire tumor regions [18]

3) Refine the final tumor border [14]

Figure 1. Flow diagram for segmentation of brain tumor Architecture

Input Image

Segmented

Output

Preprocessing

Identification

Tumor Detection

Tumor Segmentation

ROI

Seed Point

Initial Mask

Feature Extraction

Gray and White

matter

Seed Point

Normalization

Segmentation

SFCM

Morphological

Legendre Process



ISSN: 1548-7741

www.joics.org273

4. Methodology

4.1. Region Of Interest (ROI) estimation

ROI should cover the malignant tumor region without any benign tumor regions.

This method of detecting the Region Of Interest (ROI) using Spatial Fuzzy C-Means

(SFCM) algorithm is limited to a certain extend [11]. A new region growing method is

used to detect the entire tumor region even if there are more than one tumor settlements,.

The Distance Regularized Level Set Evolution (DRLSE) method is used to refine the

final and accurate tumor border [7]. Some tumors have irregular shapes and improper

boundaries. Therefore it is very difficult to identify the exact boundary of effective

algorithm that improves the accuracy of clustering. Fuzzy is a soft computing technique

C Means clustering. Fuzzy C- Means is a combined form of clustering that identifies the

tumor [8][9][11] as well as reshapes the image. The input is in one-dimensional data [3].

Fuzzy C-Means clustering performs on two-dimensional data [14]. The following

concepts are important for ROI estimation.

a) Gray Matter:

It is easy to understand the tumor as the gray color represents tumor. The Gray

Matter (GM) and White Matter (WM) are an important clinical diagnosis [10]. Gray

matter is composed of neural and glial cell, It controls the brain activity. Gray matter

consists of mostly unmyelinated neurons, most of GM are inter-neuron [15].

b) White Matter:

White Matter is made up of mostly myelinated neurons and they connect gray

matter. White Matter consists of many eliminated axons that are connected to the

cerebral cortex with other brain regions [10]. White matter is similar to gray matter [15].

It helps to identify the tumor easily.

4.2. Segmentation Of Seed Points

The Brain Tumor segmentation is a process of identifying affected tumor tissues

and protects healthy tissues from damage and identifies the tumor tissue in the brain

accurately [1][12]. Seed point detects the starting point of the tumor. Morphological

Concept is used to find out the extracted seed point[5]. Due to the irregular shape and

size of the tumor. Thus the center point of the tumor is read and clearly identified [6].

Moreover, each of initial seeds, rather than their combination should start to grow

iteratively with region growing criteria to ensure the entire tumor detection [17].

Segmentation techniques that are based on finding the regions directly.

5. Region Growing

The region growing method is an interactive image segmentation

method. The initial region begins as the extract location of the seeds. This method selects

the starting and current seed points and it is based on the location information. It is easy

to select the centroids. These techniques are generally better in noisy images. The edges

are difficult to detect. Identifying disorders and treatment planning in the field of

medicine. Edges are important features in an image to separate region.



ISSN: 1548-7741

www.joics.org274

a) Gradient

Gradient magnitude is often used to preprocess for Segmentation. Gradient helps

to identify the initial mask, flow of direction and the shape of the tumor. The level set is

dynamic. It will be moved or spread highly. It is one of the tools to identify the extract

seed point [16]. We modify the gradient image by imposing both internal and external.

The gradient shows the x and y directions.

b) Signed Distance Found (SDF)

SDF stands for Signed Distance Found. It is based on the mask to identify

distance. The Vector image may identify the length. It will be in 2- Dimensional [3].

Then it is converted into one dimensional. Legendre is used in this process.

c) Curvature

The initial shape of a tumor may be a circle, square or spherical. The curvature

will be used to identify the directions [18]. When using a gradient Two-Dimensional

image will be displayed.

d) Normalization

When the row and column are combined the Orthonormal function 2D Legendre

will be displayed. The row is read first and it is followed the column. The

Normalization shows the X and Y direction. It specifies X and Y normalization.

(Gradient rules apply).The gradient value may be divergence. Normalization shows the

intensity [2]. A clear image has been obtained by removing the blurred area of the

image.

e) Compute Divergence (Identify the Internal and External Tumor)

The level set is moved from one direction to another. It helps to identify

the starting point, X direction, Y direction and current state of the tumor [18]. Because it

is a high-grade tumor as it spread all over the brain tissues.

Algorithm

Input: MRI Images

Output: Segmented Output

1. Read the MRI Image.

2. Estimation of the Region Of Interest (ROI) using the Spatial Fuzzy

Clustering Method (SFCM) clustering results of input images.

3. Estimating the Region growth based on affinity.

4. Extract seed points using location and information of every region.

5. The similarity Method is used to determine whether the unlabeled pixels are

added to the detection region.

6. Merging the Region based on the Minimum Description Length (MDL)

criteria are used to extract non-tumor regions from the detection regions in

RGBA.

7. Apply Region Growth based Affinity (RGBA).

8. To reduce the noise and smoothing the tumor image.

9. To detect the initial counter of the tumor based on the DRLSE Method.

10. Prepare signed distance map of initial counter.



ISSN: 1548-7741

www.joics.org275

11. Requires specification of a few parameters, like the number of color band

and maximal polynomial degree for two-dimensional images.

12. Estimate the gradient descent flow for energy minimization.

13. To define X and Y Gradient and to identify X and Y direction

14. To apply the normalized gradient used to normalize the tumor image.

15. Determines convergence criteria.

16. Finally, the region is selected and segmented.

6. Result and Discussion

a) Database:

The novel Legendre Polynomial method takes medical MR image as input and

effective segmenting tumor image to make an enhanced version. The medical database

namely BRATS 2015 is a large growing database. The software testing for the proposed

Legendre polynomial method is performed by 220 cases detected as true positives are

available to make use in the following work, including 182 HGT cases and 38 LGT

cases. Exactly it covers 158 cases detected 128 in HGT and 30 in LGT. Besides 58 cases

of the remaining can be applied to evaluate our ULPA method.

b) Evaluation:

The Evaluation of the segmentations considered five metrics

Sensitivity

Dice

Specificity

positive predictive value (PPV)

Euclidean distance (ED).

Table 1. The Evaluation of five metrics

Metrics Formulas

Sensitivity

Dice

Specificity

positive predictive value

(PPV) Euclidean distance ED = (q1-p1)+(q2-p2)

The Novel Legendre algorithm performance can be evaluated in terms of

sensitivity, Dice, Specificity, PPV, and Euclidean distance ED. The Segmentation of brain

tumor defining the terms Tpos, Tneg, Fpos and Fneg from the outcome parameters and

result for the calculation of the five metrics are shown in table,



ISSN: 1548-7741

www.joics.org276

Table 2. Outcome Parameters

Outcome Parameters Ground truth

Positive Negative

Positive Detection Tpos Fpos

Negative Detection Fneg Tneg

Here Tpos is the number of Truth positives which is used to indicate the total

number of abnormal cases correctly classified. Tneg is the number of true negatives,

which is used to indicate normal cases correctly classified. Fpos is the number of false

positive and it is used to indicate wrongly detected or classified abnormal cases, when

they are actually normal cases and Fneg is the number of false negatives, it is used to

indicate wrongly classified or detected normal cases, when they are actually abnormal

cases all of the outcome parameters are calculated using the total number of samples

examined for the detection of the tumor.

Table 3. Input Case values for Truth Detection

Cases HGT LGT

220 182 32

158 128 30

Table 3 shows the tumor cases which is split it into high grade and low grade

tumor. In the first case 220 cases are split into 180 HGT and 32 LGT then 158 case set is

divided into 128 HGT and 30 LGT.

Table 4. Analysis of HGT and LGT using Truth positive

TRUTH POSITIVE (Tpos)

Grades Positive Detection Negative Detection Total

HGT 182 13 195

LGT 38 8 46

Table 4 shows the analysis on HGT and LGT using Truth Positive (Tpos)

analysis. A total of 241 cases are examined and detected as HGT and LGT. Of which

195 cases are HGT and 46 cases are LGT. HGT is split into positive and negative

detection which is accounted for 182 and 13 respectively. LGT is also split into the

positive detection with 38 cases and negative detection with 8 cases.

Figure 2. Comparison of tumor in HGT and LGT with Truth Positive (Tpos)



ISSN: 1548-7741

www.joics.org277

Figure 2 shows the comparison of tumor in HGT and LGT with Truth Positive

(Tpos). It highlights the variety in positive and negative detection.

Table 5 . Analysis of HGT and LGT using Truth Negative

TRUTH NEGATIVE (Tneg)


HGT 16 18 34

LGT 6 6 12

Table 5 shows the analysis on HGT and LGT using Truth Negative (Tneg). A

total of 46 cases are examined and detected as HGT and LGT. Of which 34 cases are

HGT and 12cases are LGT. HGT is split into positive and negative detection which is

accounted for 16 and 18 respectively. LGT is also split into the positive detection with 6

cases and negative detection with 6 cases.

Figure 3. Comparison of tumor in HGT and LGT with Truth Negative (Tneg)

Fig 3 shows the comparison of tumor in HGT and LGT with Truth Negative

(Tneg). It focuses on the variety of positive and negative detection.

Table 6 . Total Analysis of Positive and Negative detection in truth Positive

TOTAL DETECTION


HGT 198 31 229

LGT 44 14 58

Table 6 displays the total analysis of positive and negative detection. Of the 287

total detection cases , 229 cases are HGT and 58 cases are LGT. High concentration is

found in HGT with 229 cases where 198 cases are detected positive and 31 cases are

detected negative. Of the remaining 58 LGT cases, 44 cases are detected positive and 14

cases are detected negative.



ISSN: 1548-7741

www.joics.org278

Figure. 4 Comparision of total detection in HGT and LGT with Tpos and Tneg

Figure 4 compares the variation of total detection cases between positive and negative

detection

Table . 7 Input Case values for false detection

Cases HGT LGT

22 16 6

21 13 8

Table 7 shows the tumor cases which is split it into HGT and LGT. In the first 22 cases

are split into 16 HGT and 6 LGT, then 21 cases are split into 13 HGT and 8 LGT.

Table. 8 Analysis of HGT and LGT using False Positive

FALSE POSITIVE (Fpos)


HGT 12 4 16

LGT 5 1 6

Table 8 shows the analysis on HGT and LGT using False Positive (Fpos)

analysis. A total of 22 cases are examined and detected as HGT and LGT. Of which 16

cases are HGT and 6 cases are LGT. HGT is split into positive and negative detection

which is accounted for 12 and 4 respectively. LGT is also split into the positive detection

with 5 cases and negative detection with 1 case.

Figure. 5 Comparison of tumor in HGT and LGT with False Positive (Fpos)

Figure 5 shows the comparison of tumor in HGT and LGT with False Positive (Fpos). It

highlights the variation of positive and negative detection.



ISSN: 1548-7741

www.joics.org279

Table. 9 Analysis of HGT and LGT using False Negative

FALSE NEGATIVE (Fneg)


HGT 9 4 13

LGT 6 2 8

Table 9 shows the analysis on HGT and LGT using False Negative (Fneg)

analysis. A total of 21 cases are examined and detected as HGT and LGT. Of which 13

cases are HGT and 8 cases are LGT. HGT are split into positive and negative detection

which is accounted for 9 and 4 respectively. LGT are also split into the positive detection

with 6 cases and negative detection with 2 cases.

Figure. 6 Comparison of tumor in HGT and LGT with False Negative (Fneg)

Figure 6 shows the comparison of tumor in HGT and LGT with False Negative It

highlights the variety in positive and negative detection.

Table. 10 Total Analysis of Positive and Negative detection in False Negative

TOTAL DETECTION


HGT 21 8 29

LGT 11 3 14

Table 10 displays the total analysis of positive and negative detection. Of the 43 total

detection cases , 29 cases are HGT and 14 cases are LGT. High concentration is found in

HGT with 29 cases where 21 cases are detected positive and 8 cases are detected

negative. Of the remaining 14 LGT cases, 11 cases are detected positive and 3 cases are

detected negative.



ISSN: 1548-7741

www.joics.org280

Figure. 7 Comparison of total detection in HGT and LGT with Tpos and Tneg

Figure 7 compares the variation of total detection cases between positive and negative

detection

Table. 11 Existing system for Dice, PPV, Sensitivity and ED

EXISTING SYSTEM

Grades and

combination

Dice

PPV

Sensitivity

ED

HGT 0.907 0.913 0.895 2.642

LGT 0.873 0.880 0.583 2.80

Combined 1.78 0.583 1.478 5.442

Table 11 shows the Existing system for dice, PPV, Sensitivity, and ED. From an

HGT, the value for Dice (0.907), PPV (0.913), Sensitivity (0.895) and Euclidean

Distance (2.642). Similarly, from an LGT, the value for Dice, PPV, Sensitivity and ED

are 0.873, 0.880, 0.583 and 2.80 respectively. The combined value of HGT and LGT for

Dice, PPV, Sensitivity, and ED are 1.78, 0.583, 1.478 and 5.442 respectively.

Figure. 8 Comparison of Existing system for Dice, PPV, Sensitivity and ED

Figure 8 shows the comparison of Existing system for Dice, PPV, Sensitivity and ED.



ISSN: 1548-7741

www.joics.org281

Table 12. Proposed system for Dice, PPV, Sensitivity and ED

PROPOSED SYSTEM

Grades and

combination

Dice PPV Sensitivity ED

HGT 0.910 0.909 0.9128 2.980

LGT 0.880 0.877 0.8826 1.113

Combined 1.79 1.786 1.794 4.093

Table 12 shows the proposed system for dice, PPV, Sensitivity and ED

are 0.910, 0,909, 0.9128 and 2.980. Similarly from an LGT the value for Dice (0.880),

PPV (0.877), Sensitivity (0.8826) and Euclidean Distance (1.113) respectively. The

combined value of HGT and LGT for DICE, PPV, Sensitivity and ED are 1.791, 1.786,

1.794 and 4.093 respectively. Comparative study shows the proposed system is better

than the Existing system .

Figure. 9 Comparison of Proposed system for Dice, PPV, Sensitivity and ED

Experimental results

(a)

(b)

(c)

Figure. 9 (a) Input Image (b) Gray Matter (c) White Matter



ISSN: 1548-7741

www.joics.org282

Region growing

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

Figure. 10 Experimental result for the spatial fuzzy C means clustering method to evaluate Region

Of Interest (ROI) (a) Input MRI image (b)-(g) is the levels of ROI (h) Segmented tumor region.

The Spatial Fuzzy C Means clustering can be evaluated region growing which is overcome the

method Unified Legendre Polynomial. The result of the Unified algorithm is

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

Figure. 11 Experimental result for the Legendre Polynomial method to evaluate Segmentation (a)

Input MRI image (b)-(g) is the levels of Legendre (h) segmented tumor region.



ISSN: 1548-7741

www.joics.org283

7. Conclusion

In this paper we propose a Unified Legendre Polynomial algorithm for tumor

segmentation named ULPA. We use SFCM clustering to justify the medical image

segmentation and to find out ROI depending on its location. The seed point is expanded

to grow independently based on the affinity method. The spatial distance is between the

neighboring pixels and the region growing. The method is a way forward to refine the

result of region growing, gradient and normalization. The Experimental results

demonstrate the effectiveness of our ULPA in MRI images.

References

[1] Maddalena Strumia et al “White Matter MS-Lesion Segmentation Using a Geometric

Brain Model” IEEE Transactions on Medical Imaging, Vol. 35, (2016), PP. 1636 –

1646.

[2]. Sergio Pereira et al “Brain Tumor Segmentation using Convolutional Neural

Networks in MRI Images” IEEE Transactions on Medical Imaging, Vol. 35, No. 5,

(2016), PP. 1240-1256.

[3]. Shang-Ling Jui et al” . Brain MR Image Tumor Segmentation with 3-Dimensional

Intracranial Structure Deformation Features” IEEE Intelligent Systems, Vol. 31, (2015),

PP.66-77.

[4]. Bjoern H. Menze et al” A generative probabilistic model and discriminative

extensions for brain lesion segmentation – with application to tumor and stroke” IEEE

Transactions on Medical Imaging, (2015), PP.933-946.

[5]. Annemie Ribbens, et al“Unsupervised Segmentation, Clustering and Groupwise

Registration of Heterogeneous Populations of Brain MR Images” IEEE Transactions On

Medical Imaging” Vol. 33, Issu. 2, (2014),PP. 201-224.

[6]. Colm Elliott et al “Temporally Consistent Probabilistic Detection of New Multiple

Sclerosis Lesions in Brain MRI” IEEE Transactions On Medical Imaging, Vol. 32, Issue.

8, (2013).

[7]. Zexuan Ji et al” Fuzzy Local Gaussian Mixture Model for Brain MR Image

Segmentation”IEEE Transactions On Information Technology In Biomedicine, Vol. 16,

Issue. 3,(2012).

[8]. Y.Chen1 J. Zhang, S. Wang, Y. Zheng ”Brain magnetic resonance image

segmentation based on an adapted non-local fuzzy c-means method”, IET Computer

Vision,Vol. 6, Iss. 6,(2012), pp. 610 –625.

[9] A. Ortiz et al” MR brain image segmentation by growing hierarchical SOM and

probability clustering” Electronics Letters, Vol. 47 Issue. 11,(2011).

[10]Yutaka Hata et al “Automated Segmentation of Human Brain MR

Images Aided by Fuzzy Information Granulation and Fuzzy Inference” . IEEE

Transactions On Systems, Man, And Cybernetics—Part C: Applications And Reviews,

Vol. 30, Issue. 3, (2000).

[11] Shan Shen, William Sandham et al, ”MRI Fuzzy Segmentation of Brain Tissue using

Neighborhood attraction with neural networks optimization” IEEE Transactions on

Information Technology in Biomedicine, Vol. 9, No.3 (2005), PP. 459-467.



ISSN: 1548-7741

www.joics.org284

[12] Fuyong xing et al, “An Automatic Learning Based framework for Robust Nucleus

segmentation” IEEE Transaction, “Multireference Level Set for the characterization of

Nuclear morphology on Medical Imaging, vol 35, No 2, Feb (2016), PP. 550-556.

[13] Youngyue zhang et al, “Segmentation of Brain MR Images through a Hidden

markov Random field model and the expectation Maximization Algorithm” IEEE

Transaction on Medical Imaging, Vol 20, Issue 1,(2005), PP 45-57.

[14] S.MathuKumar and N.Santhiyakumari, “Evaluation of K-Means and Fuzzy C-

Means Segmentation and MR Images of brain”, Egyptian Journal of Radiology and

nuclear medicine Vol. 46, Issue. 2, (2015), PP 475-479.

[15] E.A. Zanaty et al, “Determination of Gray Matter (GM) and White Matter (WM)

volume in brain Magnetic Resonance Image (MRI)”, international Journal of

computer Applications, Vol.45,(2012), PP. 16-22.

[16] S.N. Deepa and B. Arunadevi, “ Extreme Learning Machine for classification of

Brain tumor in 3D MR Images “, Informatologia, Vol. 46, no.2,(2013), PP.111-121.

[17] Aastha sehgal, Shashwat Geol, Parthasarathi Mangipudi, “Automatic Brain

Tumor Segmentation and Extraction in MR Images” Conference on Advances in Signal

Processing (CASP), Pune , 2016.

[18] Nilesh Bhaskarran Bahadure, Arun kumar Ray and Har pal Thethi, “Image

analysis for MRI Based brain Tumor detection and feature Extraction using biologically

inspired BWT and SVM”, International Journal of Biomedical imaging, Article ID

9749108, (2017), PP. 12 – 26.

[19] N.Gordillo, E.Montsency and P.Sobrevilla, “State of the art survey on MRI brain

tumor segmentation”, Magnetic Resonance Imaging, Vol. 31, no.8, (2013), PP. 1426-

1438.



ISSN: 1548-7741

www.joics.org285

Brain Tumor Segmentation Using A Novel Unified Legendra ...joics.org/gallery/ics-1164.pdf · The Brain Tumor segmentation is a process of identifying affected tumor tissues and protects

Documents