Page 1
http://www.iaeme.com/IJARET/index.asp 44 [email protected]
International Journal of Advanced Research in Engineering and Technology (IJARET) Volume 11, Issue 4, April 2020, pp. 44-56, Article ID: IJARET_11_04_006
Available online athttp://www.iaeme.com/IJARET/issues.asp?JType=IJARET&VType=11&IType=4
ISSN Print: 0976-6480 and ISSN Online: 0976-6499
© IAEME Publication Scopus Indexed
AN IMPROVED IDENTIFICATION METHOD OF
ABDOMINAL AORTIC ANEURYSM USING
EXUDATE SEGMENTATION AND DNN
S. Anandh
Research Scholar, Department of Biomedical Engineering, Bharath University, Chennai,
Tamilnadu, India.
Dr. R. Vasuki
Professor and Head, Department of Biomedical Engineering, Bharath University, Chennai,
Tamilnadu, India.
Dr. Y. Premila Rachelin
Assistant Professor, Department of Physics, Scott Christian College, Nagercoil, Tamilnadu,
India.
ABSTRACT
Changing over the progressing headway of AAA (abdominal aortic aneurysm)
advancement and revamping data for judicious treatment needs a noteworthy and
computational point of view exhibition framework. As AAA is deadly and rupture
disease a powerful treatment is required. An examination setup is built up in this
research to calculate the centers around the precise discovery of the AAA picture. The
info AAA data is preprocessed to convert the RGB scale into gray scale pictures by
utilizing versatile middle channel, additionally the pixels which are adulterated by
excessive clamor is also decided. And then exudates based division method is applied
before removing the unnecessary featured component from AAA pictures. After the
extrication, the best highlights are chosen by utilizing spatial kernel FCM. Lastly so
as to order and acknowledgment, profound neural classifier is applied. The proposed
framework achieves our point in predicting the AAA progress and in calculating the
proliferation weakness. The presentation of our framework is estimated by utilizing
exactness, accuracy, f-score and calculation time are used. The examination results
demonstrate the superior presentation of the proposed scheme over the current
classifiers.
Key words: DNN Classifier, Fuzzy C-means Clustering, Exudates segmentation,
Adaptive Median filter.
Page 2
S. Anandh, Dr. R. Vasuki and Dr. Y. Premila Rachelin
http://www.iaeme.com/IJARET/index.asp 45 [email protected]
Cite this Article: S. Anandh, Dr. R. Vasuki and Dr. Y. Premila Rachelin, an
Improved Identification Method of Abdominal Aortic Aneurysm Using Exudate
Segmentation and Dnn, International Journal of Advanced Research in Engineering
and Technology (IJARET), 11(4), 2020, pp 44-56.
http://www.iaeme.com/IJARET/issues.asp?JType=IJARET&VType=11&IType=4
1. INTRODUCTION
AAA is a vascular sickness emerges because of degenerative, proactive, mycosis infection
and furthermore causes arteriosclerosis. It brings about widening of the abdominal aorta due
to weakened aortic cells. The untreated AAA will extend various events growing the risk of
aortic satisfaction. These aneurysms impact around 0.3% of people with more than 60 year of
old. Half of these aneurysms are recognized by the end for about 10% of the influenced
patients will be broken as a result of weakened vein divider. The savage result is nearly 80%
of patients with split aneurysms will exist no more. The various decisions of treatment
accessible nowadays are nosy technique is suggested considering endovascular circumstance
of aortic joins via an unimportantly prominent place on the body of the patient. In order to
implement the procedure successfully, the fitting stunt joins contraption must be picked.
Remembering the ultimate objective to choose the stunt join of legitimate size and shape,
careful information on aortic structure and measurement are required. The helpful picture
preparing methodologies sought after by fitting picture examination systems have seemed, by
all accounts, to be significant for the estimation of abdominal aneurysms.
Heart disease, explicitly aortic aneurysm is the 13th driving infection behind the death
toll. There are few couple of treatment decisions accessible for AAA. One of the options is
space clinical methodology where the patients are left free at the medical zone and a stunt is
sunk in the vascular prosthesis tissues. This sort of mediation is uncommonly unsavory to the
patients and not suit for everyone. Another decision is endovascular stunt game plan.
Expansion of this stunt is done by a catheter strategy generally via small cut in the femoral
course. The aneurysm expansiveness is the basic fundamental point if the intervention ought
to occur. In the event that there ought to be a situation of AAA, an attempt is made to balance
the estimation with the vessel core in excess of 55mm. Right now, several methods are
devoted to assist doctors with respect to the AAA segmentation. A self-loaded isoperimetric
computation for the segmentation of AAA is used.
A working setup exhibits in similar manner has been used for the segmentation of AAA.
The author Dale .M proposed a familiar technique with the partition inside and outer state of
thyroid aortic aneurysm with a working structure showing a basic to keep the shape at a
particular detachment [1]. Hahn et al showed a procedure to piece rounded structure
considering active visual mode and associated with the segmentation of AAA [2]. Particular
responses regarding the segmentation of aorta are found in engraving. The fused aortic center
point removal and periphery computation, artificial intelligence frameworks, zone creating or
watershed & level set techniques [3]. Behrens delineated a Hough transformation based
figuring got together with kalman filters. The burden of these strategies was, regardless, the
prerequisite for 3 customers picked variables specifically, starting phases of the aortic, the
aortic length, and harsh center bearings. He developed a totally modified division procedure
for the aorta, using a Hough transformation based deformable surface mode innovation.
Conversely in Behrens strategy, the proposed method join dynamically from the prior
anatomic getting the hang of in regards to the condition of the aorta bend and along these lines
no manual involvement is needed [4].
In order to level the CTA pictures, an enemy of structural scattering strategy was utilized.
The Lumens are obtained from the pre-prepared picture by 2 stages via division and
Page 3
An Improved Identification Method of Abdominal Aortic Aneurysm Using Exudate Segmentation
and Dnn
http://www.iaeme.com/IJARET/index.asp 46 [email protected]
morphological activity. Imprisonment of the stomach lumen is built up by using structural
information, logical morphology and grouping experiment. It is hard to fragment the internal
and external mass of the abdominal aneurysm district. Henceforth the precise division and
estimation of aneurysm is performed in [5].
To distinguish the thrombus in the cardiac framework, 3D level set and anisotropic
dispersion are utilized and 3-dimensional CTA picture information is applied. The
geometrical deformed model, for example, dynamic shapes strategies relying upon area or
edge by using fractional unmistakable condition executed by level set. The Courant Friedrichs
Lewy (CFL) is a soft equation for expressing Finite Distinctive Method for following bends.
For portraying and lessening the associations at delicate limit, a dispersion work has been
utilized [6].
The primary estimations of the required stunt up and comer have been considered for
fitting, estimations from the pictures. Programmed identification and division is
acknowledged from 3D characteristic of the CT. Initial luminal community is distinguished
for division. The estimations of width are in vertical with the vessel bearing and a genuine
three dimensional way extend of the lumen is trailed by ESM for every coordinated locale [7].
In the current framework we present another strategy called exudate division for
acquiring exact limit esteems with bends. Contrasted and the current strategies this division
model takes out the re-initialization issues. It is actualized in certain territories like clinical
picture handling, remote detecting and in automated vision. In loud pictures the disadvantage
is decreased and calculation time is diminished utilizing the middle channel and spatial
portion fluffy c implies grouping. First the clamor is decreased by applying versatile middle
channel at that point exudates division is utilized so as to get the de noised picture. At that
point spatial piece fluffy c implies bunching is applied to separate the item limit utilizing
post-division technique. At last profound neural system is used for conclusive
characterization. The accuracy and the deliberateness of the proposed strategy for different
stomach aortic pictures are portrayed. Consequently the test shows that it is extremely
effective technique for the loud pictures and exact division is applied.
2. PREVIOUS WORKS
A few existing methodologies that were portrayed before are given beneath:
Roy .D et al [8], presented a repetitive model-constrained graph slice estimation to
segment AAA. Given the fundamental aortic lumens segmentation, the presented method
subsequently partitions the aneurysm by over and again incorporated power subordinate graph
min-split segment and geometric parametrical model fitting. The geometrical model can
effectively constrain the graph min-cut segment from "spilling" to near to veins and organs.
Elaborated results on aneurysm CT data gives healthy segments of the thrombus aneurysm
rapidly time with a normal by and large size complexity of 10% and normal volume spread
error of nearly 12% with basically indistinguishable from the cover passerby botch.
Pham T.D et al [9] developed the new method as volume rendering. It is a noteworthy and
basic framework for legitimate recognition. One without a doubt comprehended application
locale is the proliferation and portrayal of output from restorative scanning like pictured
tomography. 2D dark worth cuts made by the scanning device are replicated and appeared as
a three dimensional show. Structural view of therapeutic picture should discover 2 basic
problems. For the essential spot, it is difficult to partition remedial yields as individual
element considering power regards. This brings about density that contains a ton of
insignificant or unnecessary material. Then, regardless of the way the dark scale picture are
the normal method for demonstrating remedial volume; these types of pictures are not so
Page 4
S. Anandh, Dr. R. Vasuki and Dr. Y. Premila Rachelin
http://www.iaeme.com/IJARET/index.asp 47 [email protected]
much fitting for including zones of energy within the volume. Examinations of the pragmatic
system exhibited that every one of the force regard are difficult to distinguish in a dim scale
pictures. In that condition, concealing is a lot of feasible visual component, as a low stage
video system could rapidly and absolutely distinguish the closeness or isolation of a particular
objective concealing in a different toned pictures.
Sassani S.G et al [10] showed another procedure for the area and change of segment spills
in volume therapeutic picture. Divisional spills occur if the segment volume develops outside
the genuine objective breaking point as neighboring structures. This strategy recognizes the
segment spill premise limited by point size likelihood portrayal, fits a surface up to the break
limit, and segregates the gap from the target structure in the way of identifying the essential
cutoff. Our methodology is modified, doesn't rely upon prior information, and is liberated
from the fundamental division strategy. Exploratory evaluation on 1450 division of aortic
aneurysm from topographical channels by utilizing essential segment techniques yield a
difference in 55% (standard=14.5) and 49% (standard=19.4) in the normal surface partition
and the volume spread screw up the fundamental and the helped division.
Bahrendt et al [11] proposed a method for segmenting the outer level of AAA in
tomography obtaining was shown. Our procedure starts by enrolling the lumen centreline
among the customers portrayed seed centers that fill in as cutoff points to the injury. The
midpoint chooses the several Planar Reconstruction planes which depicts the vessel cross
breed portions. The customer genuinely locates the epithelial area of the essential proximal
planes for resizing the structures. In the beginning, a round is made: the structure is used to
frame a kept interest area in the neighboring separation planes by means of the centre line.
The voxel of the bordering planes inside the interested region and the voxel on which the
shape suits were used to draw an organized graph. By then, the min cut of the graph was
registered, thus making a perfect segment of the aneurysm on the close by planes. This
divided shape completes in as a starting level for the circle, which is implemented over once
more.
Andrzej et al [12] presented a strategy for aneurysm. It is a local growth of the Aortic
valve which occurs between the renal and iliac passageways. The incapacitating of the aorta
prompts its mutilation and creates thrombus. Starting then, this technique uses medicines to
incorporate the option of an endovascular prosthetic, the advantage of which is unimportant
meddling procedure, yet also anticipates that watching should separate post-operative
outcome. So as to satisfactorily overview the movements experienced after clinical method, it
is imperative to area the AAA, which is an extraordinarily repetitive task. Here we delineate
the fundamental outcomes of another powerful learning blend plot for the self-loaded ID and
segment of the lumen and the AAA, which uses pictures power features and discriminative
classifiers.
S. Habib et al [13] developed a cross breed deformable displaying procedure for
fragmenting the AAA utilizing NURBS location which works on 2 variables mapping to a
surfaces in 3D spaces. B-Spline location is utilized to structure the mishappening
consequently; it provides more grounded smoothness limitations un-uniform appropriation.
Informational indexes are huge in shape; so best execution is come to by this twisting
displaying. A least estimation is required as a result of the minimum measures of working
focuses. NURBS smoothness restrictions of the segment maintain a strategic distance from
spillage in the close by locale. Most exact outcomes are acquired by this calculation. The
minor post handling is adequate which is outwardly assessed by the clinical specialists. In any
case, the division result precision isn't gotten. Parting and converging of sizes of our twisting
model ought to be permitted. Troublesome bifurcation position is a significant disadvantage
of this model.
Page 5
An Improved Identification Method of Abdominal Aortic Aneurysm Using Exudate Segmentation
and Dnn
http://www.iaeme.com/IJARET/index.asp 48 [email protected]
B.B. Nakhjavanlo et al [14] presented another methodology of a nonlinear powerful
displaying classification of abdominal aneurysm. Picture marker is utilized for the assurance
and the executives of crack hazard treatment that needs no exact picture division. Estimating
the consistency of picture signal is the fundamental idea of this procedure. Nonlinear
powerful model joined with shrouded Markov model was created to evacuate imaging marker
for threat forecast. Testing of increasingly comprehensive information is required as the
primary disservice of this technique.
3. RESEARCH METHODOLOGY
In the proposed work, the size and position are identified as critical parameter for the
specialists so as to give a superior treatment. The principle goal of this approach is to upgrade
the precision. During preprocessing, versatile middle channel is utilized to decide the pixels
that are influenced by the clamor. Before doing the division in AAA picture, the first picture
must be changed in to dark scale picture. Division is the way toward partitioning the picture
into the homogenous area. It's a difficult assignment. Division is finished by utilizing the
watershed change. It performs the activity by utilizing separation change, slope strategy and
marker extraction. In the wake of performing division, highlights are extricated by fluffy c-
implies bunching. By planning a viable wellness work, it produces high precision in
characterization. At long last deep neural network is utilized for the arrangement presented in
Fig.1. It ranges across at least 2 layers and it is repeated for multiple times to create a high
precise outcome.
Figure 1: Block diagram of proposed method
Page 6
S. Anandh, Dr. R. Vasuki and Dr. Y. Premila Rachelin
http://www.iaeme.com/IJARET/index.asp 49 [email protected]
3.1. ADAPTIVE MEDIAN FILTER:
It assists in finding the pixels in a picture that are tainted by imprudent commotion as
presented in Fig.2. It orders the pixels as clamor by comparing with every other pixel in a
picture. It's a capable model for upgrading the nature of pictures. In versatile middle sifting,
new calculation has been suggested to safeguard the sharpness and to discover the structure of
the rash commotion. This channel is estimated as far as sign to commotion proportion and
time productivity. In versatile middle channel, in first cycle it identifies the situation of ruined
uproarious pixel. The twofold worth zero and one is utilized to discover the loud pixel. Zero
demonstrates the pixel is acceptable and one speaks to the boisterous pixel. This channel
expels the whole commotion pixel.
Let Ii, j speaks to the pixel of boisterous picture, I min speaks to the base pixel worth and I
max shows the greatest pixel esteem.
Level1:
If Imin< Imed <Imax, at that point the middle worth isn't a motivation, so it changes to stage 2
to identify if the present pixel is a drive. Or the size of the window gets most extreme and
stage 1 is repeated until the middle worth isn't a motivation so the calculation goes to level 2;
or the greatest window size is come to
Level2:
If Imin<Ii,j<Imax, at that point the present pixel esteem isn't a motivation, so the separated
pixel of a picture stays unaltered. Else the pixel estimation of a picture is either equivalent to
Imax or Imin. The separated pixel of a picture is relegated as middle an incentive from stage
1.
Figure 2: Flowchart of adaptive median filter
3.2. Exudate Segmentation:
It's a difficult errand to perform division in AAA since it has high pixel similitude to
contiguous tissue. Division's principle aim is to order the picture into different segments every
one of which demonstrates different informations in the picture, for example, shading, force
or surface. Division approach rely upon two properties,
Page 7
An Improved Identification Method of Abdominal Aortic Aneurysm Using Exudate Segmentation
and Dnn
http://www.iaeme.com/IJARET/index.asp 50 [email protected]
Distinguishing discontinuities and
Similitude
In PC supported analysis, programmed exudates division is a significant undertaking. The
primary commitment of this is division of exudation from aneurysm picture by utilizing
versatile limit. After the procedure of histogram, a versatile limit is chosen by utilizing initial
request measurable variables, for example, average and standard deviation. Histogram is
developed by dividing the picture into equivalent estimated classes shown in Fig.3. At that
point for every class, the quantity of focuses from the data which lies under every class is
checked.
Vertical pivot: Frequency
Flat pivot: reaction variable
Figure 3: Representation of Histogram
Versatile edge is proceeded as follows,
1. Pick an underlying assessment
2. Divide the dark scale picture utilizing T. It creates two gatherings of pixels.
3. S1 is comprises of pixels with values more noteworthy than T.
4. S2 comprises of pixels with values not as much as T.
5. Compute the normal dark level worth m1 and m2 for the pixels which is available
in S1 and S2.
6. Compute the new edge (T) esteem by utilizing the condition,
( ) (1)
7. Emphasize the even stages till distinction in T in ceaseless cycle is littler than a
predefined variable p0.
This strategy is proficient and sets aside a limited capacity to focus effort for performing
division of exudates in a continuous circumstance. The abdomen aneurysm picture is changed
over to dark scale picture and experiences the procedure of histogram. It is broke down and
utilized to decide a reasonable limit for division of exudates.
The limit utilized for portioning the exudate is as per the following,
[{ ( )) ( )} ( ) (2)
Where Th speaks to the versatile limit
I demonstrate the standardized dim shading channel
Page 8
S. Anandh, Dr. R. Vasuki and Dr. Y. Premila Rachelin
http://www.iaeme.com/IJARET/index.asp 51 [email protected]
From the limit parallel picture, bogus positive pixels were expelled for the precise division
of exudates.
3.3. SPATIAL KERNEL FCM (SKFCM)
In picture handling gathering and division are nearest terms. Pixels with comparable
characteristics are part contingent upon the factors like spatial information and separation
gauges in the bunching strategies. The aftereffect of the gathering procedure is portrayed in
the spatial space as different regions for division process. A few likeness metric like interim,
proclivity, sufficiency are isolated in bunching. Spatial FCM (SFCM) was acquainted all
together with decrease the clamors and exceptions however less measure of commotion is
evacuated. Spatial Kernel FCM Clustering (SKFC) is acquainted to give better and quicker
division result contrast with SFCM and FCM. Utilizing ordinary FCM strategy, SKFCM
coordinates spatial information and uses Gaussian RBF bit administrator. It provides best
outcome in various fields.
3.3.1. SKFC ALGORITHM
There are 6 stages:
Dole out the picked pixels of the dataset "X" and fix the middle worth Ɛ with m as
weighting example for each fluffy participation work.
Sponsorship esteems are figured against each focal point of the pixels with the
end goal that.
= ( ( ))
( )⁄
∑ ( ( ))
( )⁄
calculate new centre values
=∑
∑
is a new membership esteem is obtained by,
=
∑
= ∑ ( )
Here NK( ) is a block window lied in the spatial domain
J is Objective function is obtained by,
∑∑
( ( ))
The condition {J(i)-J(i=1)}<ε for limit of end is assessed with end condition as Ɛ.
Stop the methodology if the criteria is fulfilled , in any case rehash from the
second stage till it fulfil the condition.
3.4. DEEP NEURAL NETWORK:
It's a neural system with a few layer of hubs among info and yield. For order and
acknowledgment, profound neural system classifier is utilized, it's anything but a basic layer
of calculation presented in Fig.4. It ranges across at least two layers. The three layers included
are
Page 9
An Improved Identification Method of Abdominal Aortic Aneurysm Using Exudate Segmentation
and Dnn
http://www.iaeme.com/IJARET/index.asp 52 [email protected]
(1) Source layer
(2) Hidden layer and
(3) Yield layer.
In the event that f(x) is nonlinear, a system with 1 shrouded layer can play out any
arrangement issue. A lot of weight present can deliver the objectives from the info. While
compared with ANN, neural systems utilize nonlinear F(x) so they draw complex limits yet
they keep the information unaltered. Neural system design ought to be fit for learning the
genuine basic highlights hence it sums up quite well. It plays out a noteworthy activity when
the issue turns out to be increasingly confused. So as to perform arrangement, there is a need
to prepare a lot of information. It aids in exact forecast of aneurysm picture.
Figure 4: Neural network layers
(1)The preparing information and the comparing marks are provided to the model.
(2)The whole data will be forwarded to the model and it is iterated for multiple times and
in every cycle it has 256 stages.
(3)At last it creates the yield.
4. RESULTS
A versatile middle channel is used for surface evaluation to propose whether there are specific
recurrent substance exists in the picture explicitly course in an obliged locale around the
purpose of assessment. From the channel, the yield MRI AAA picture is isolated.
Watershed division is the least irksome division procedure used to choose the lower level
of pixel. The foundation and the difference are the 2 essentialities. The diminish level has the
vast majority of the substance in the picture. Right now, dull level is immensely pinnacle. By
watching the specific pictures, the observer can decide overall tonal scattering from the start.
As the information contained in the structure is a depiction of pixel movement as a section of
tonal assortment, picture is separated into pinnacle or valley.
An inherited estimation is introduced to improve the picture division and researches the
strategy space by a technique that s lopsided as each pixel is accumulated. Hereditary count is
the sensibly improved framework. It is important in picture advancement and division. It is a
noteworthy wide arrangement space. This explains the growing notoriety of GAs application
in picture preparing field. Our proposed work is implemented in Matlab.
Page 10
S. Anandh, Dr. R. Vasuki and Dr. Y. Premila Rachelin
http://www.iaeme.com/IJARET/index.asp 53 [email protected]
Figure. 5 Input image
The Fig.6 speaks to the AAA picture that has both haze and radiation commotions which
are then expelled by versatile middle channel.
Figure. 6 Noise reduced image
Figure. 7 Binary patterned image
Figure. 8 Segmentation result from the input image
Processing of information standardizes pictures. The division and requesting of restorative
aneurysm pictures allows us to recognize zone and thickness seeing by using the district
expelled continuously condition. This assessment examines the believability of tumor
affirmation using pixel (seed) point features for perceiving the affected locale in the cell.
The delayed consequences of aneurysm division and the affirmation of human and AI
computations are contrasted and assessed with demonstrate our execution better than the other
feature based division and ID.
Page 11
An Improved Identification Method of Abdominal Aortic Aneurysm Using Exudate Segmentation
and Dnn
http://www.iaeme.com/IJARET/index.asp 54 [email protected]
4.1. Performance parameters comparison
Figure 9: Accuracy
Figure 10: Sensitivity
Figure 11: Specificity
Table 1. Comparison of Accuracy and precision
IMAGES
ACCURACY PRECISION
WITHOUT DNN
CLASSIFIER
WITH DNN
CLASSIFIER
WITHOUT DNN
CLASSIFIER
WITH DNN
CLASSIFIER
1 88.2 89.4 89.4 90.8
2 86.6 87.2 86.2 87.9
3 87.5 88.4 87.3 88.2
4 90.3 92.6 90.1 91.5
5 89.7 91.7 90.7 92.3
Page 12
S. Anandh, Dr. R. Vasuki and Dr. Y. Premila Rachelin
http://www.iaeme.com/IJARET/index.asp 55 [email protected]
Table 2. Comparison F-Score and computational time
IMAGES
F- SCORE COMPUTATION TIME(ns)
WITHOUT DNN
CLASSIFIER
WITH DNN
CLASSIFIER
WITHOUT DNN
CLASSIFIER
WITH DNN
CLASSIFIER
1 84.3 86.3 0.83 0.82
2 82.9 83.8 0.84 0.82
3 88.4 90.8 0.85 0.83
4 87.2 88.7 0.82 0.80
5 85.7 87.5 0.85 0.81
The above table 1 and 2 gives the deviation in requesting of MRI AAA pictures with
single level division and leading with the comparable pictures with our proposed watershed
division.
5. CONCLUSION
An Exudate division strategy is proposed right now distinguish and portion the image
obtained from MRI Abdominal aorta area. In regular strategy, re-initialization issues are high.
Re-introduction issue is completely wiped out in the proposed technique. Versatile middle
channel decreases the clamour in the picture productively. Related with standard Fuzzy C
Means calculation, profound neural system (DNN) is using the Gaussian RBF part utility as a
separation variable by consolidating spatial information. This technique is diverged from the
ongoing extrication strategies. The exhibition is assessed as far as Accuracy, coefficient,
generally speaking introduction and mistake rate and clamour.
REFERENCES
[1] Dale, M. A, Suh, M. K, Zhao, S., Meisinger, T, Gu, L, Swier, V. J.,Xiong, W, Background
differences in baseline and stimulated MMP levels influence abdominal aortic aneurysm
susceptibility, Atherosclerosis, 243(2), 2015, pp.621–629.
[2] Hahn, S., Morris, C. S., Bertges, D. J., & Wshah, S., Deep Learning for Recognition of
Endoleak after Endovascular Abdominal Aortic Aneurysm Repair, International Symposium
on Biomedical Imaging, 3, 2019, pp.1-8.
[3] Marleen de Bruijne, Wiro J. Niessen, Associate Member, J. B. Antoine Maintz, and Max A.
Viergever, Localization and Segmentation of Aortic Endografts Using Marker Detection,
IEEE Transactions On Medical Imaging, 22(4), 2013, pp. 276-283.
[4] Behrens T, Rohr K, Stiehl HS, Robust segmentation of tubular structures in 3-D medical
images by parametric object detection and tracking, IEEE Transactions on Systems, Man, and
Cybernetics, 2003, 33(4), pp. 554-61.
[5] Roger C.Tam, Christopher G.Healey, Borys Flak and peter Cahoon, Volume Rendering of
Abdominal Aortic Aneurysms, IEEE Visualization, 97(6), 2015, pp. 712-729.
[6] M. Auer and T. Christian Gasser, Reconstruction and Finite Element Mesh Generation of
Abdominal Aortic Aneurysms from Computerized Tomography Angiography Data with
Minimal User Interactions, IEEE Transactions on Medical Imaging, 29(4), 2016, pp. 654-668.
[7] Do, H. N, Ijaz, A, Gharahi, H, Zambrano, B, Choi, J, Lee, W, & Baek, S, Prediction of
Abdominal Aortic Aneurysm Growth using Dynamical Gaussian Process Implicit Surface,
IEEE Transactions on Biomedical Engineering, 66(3) , 2018, pp. 609 – 622.
[8] Roy, D, Holzapfel, G. A, Kauffmann, C, Soulez, G, Finite element analysis of abdominal
aortic aneurysms: geometrical and structural reconstruction with application of an anisotropic
material model, IMA Journal of Applied Mathematics, 79(5), 2014, pp. 1011–1026.
Page 13
An Improved Identification Method of Abdominal Aortic Aneurysm Using Exudate Segmentation
and Dnn
http://www.iaeme.com/IJARET/index.asp 56 [email protected]
[9] Pham, T. D. Golledge, J, Pattern analysis of imaging markers in abdominal aortic aneurysms,
International Biomedical Engineering and Informatics, 66(3), 2013, pp. 609 – 622.
[10] Sassani, S. G, Kakisis, J, Tsangaris, S, & Sokolis, D. P, Layer-dependent wall properties of
abdominal aortic aneurysms: Experimental study and material characterization, Journal of the
Mechanical Behavior of Biomedical Materials, 49, 2015, pp.141–161.
[11] Behrendt, C.A, Dayama, A, Debus, E. S, Heidemann, F, Matolo, N. M., Kolbel, T, &
Tsilimparis, N., Lower Extremity Ischemia after Abdominal Aortic Aneurysm Repair, Annals
of Vascular Surgery, 45, 2017, pp.206–212.
[12] Andrzej Polanczyk, Michal Podgorski, Maciej Polanczyk, Aleksandra Piechota-Polanczyk,
Christoph Neumayer, Ludomir Stefanczyk, A Novel Patient-Specific Human Cardiovascular
System Phantom (HCSP) for Reconstructions of Pulsatile Blood Hemodynamic Inside
Abdominal Aortic Aneurysm, IEEE Access, 6, 2018, pp. 61896 – 61903.
[13] S. Habib, J. Dehmeshki, Automatic segmentation of Abdominal Aortic Aneurysm, IEEE
transactions on CSIT, 2018, pp. 412-415.
[14] B. B. Nakhjavanlo, T. J. Ellis, P.H.Soan, J. Dehmeshki, 3D Medical Image Segmentation
Using Level Set Models and Anisotropic Diffusion, Seventh International Conference on
Signal Image Technology & Internet-Based Systems, 2011, pp. 403-408.
[15] Sreedhar T and Dr. Sathappan S, A Comparative Analysis of Image Segmentation Techniques.
International Journal of Computer Engineering and Technology, 9(5), 2018, pp. 64-69.
[16] Noor Adnan Ibraheem and RafiqulZaman Khan, A Research Study on Recent Skin Color
Based Statistical Segmentation Modeling Techniques, International Journal of Graphics and
Multimedia (IJGM), Volume 6, Issue 1, January-April- 2015, pp. 01-07
[17] Amel H. Abbas, Aryan A. Kareem and Mohammed Y. Kamil, Breast Cancer Image
Segmentation Using Morphological Operations, International Journal of Electronics and
Communication Engineering & Technology (IJECET), Volume 6, Issue 4, April (2015), pp.
08-14
[18] Dr. Ashwin Patani, Color-Segmentation of Fabric Images, International Journal of Electronics
and Communication Engineering & Technology (IJECET), Volume 5, Issue 5, May (2014),
pp. 64-79