ISSN 2278-3091

Volume 9, No.3, May - June 2020

International Journal of Advanced Trends in Computer Science and Engineering
Available Online at http://www.warse.org/IJATCSE/static/pdf/file/ijatcse207932020.pdf
https://doi.org/10.30534/ijatcse/2020/207932020

Classification of Coronary Heart Artery Disease Using IVUS

Images by SVM Classifier with Modified Radial Basis
Function Kernel (MRBFK)
S. Sridevi1, M. Sundaresan2
1
Research Scholar, Department of Information Technology, Bharathiar University, Coimbatore, India
devisris.mphil@gmail.com
2
Professor and Head, Department of Information Technology, Bharathiar University, Coimbatore, India
bu.sundaresan@gmail.com

 allocate, demonstrate and store the patient data [1]. The

ABSTRACT increasing volume of medical data demands new techniques
for organizing, sharing, managing and extracting knowledge.
In real world phenomenon, Digital image processing
techniques and algorithms are widely used in the medical The knowledge extracted can be further used for diagnosing
applications. The medical applications are used to predict the disease, preventing human error and informing about the
disease for reducing the death rate. But, nowadays heart functions and consequences of several diseases. Moreover,
disease is one of the most significant and exigent problem. advances in digital image processing system and more
Based on the problem, heart disease can be in different form specifically, image analysis takes a major hand in medical
like atherosclerotic plaque, calcium accumulation, and vice progress, as well as analysis of medical images play an
versa. Here this paper concentrates only on atherosclerotic ever-increasing crucial role in terms of research and medical
plaque, which is the major cause of coronary heart artery diagnosis [2]-[4].
diseases and human deaths. Coronary heart artery plaque
layer analysis is one of the most important and challenging Coronary Angiography is a standard method for medical
problems in real world phenomenon. For this purpose, this imaging procedures and diagnosis of coronary heart disease [5].
paper elucidates and presents a five phased approach of (1) Angiogram technology does not provide the heart artery layer
Cartesian to Polar (C2P) Coordinate conversion (2) Iterative and vessel wall information for diagnostic procedure. It is mainly
level Energy Computation Method (IECM) (3) Feature failed to assess the plaque layer boundary, which is responsible
Extraction (4) Classification and (5) Performance Evaluation. for partial or total obstruction of the arteries [6]. Nowadays,
Hence, the performance is analyzed for classification of IVUS Intravascular Ultrasound (IVUS) has been introduced as an
images based on Sensitivity, Specificity and Precision. The alternative technique to angiography diagnostic aiming at
proposed method achieves Accuracy of 88% and the Kappa more accurate imaging of coronary atherosclerosis. The
value of 0.92 which indicates the good agreement test using specific feature extraction is necessary in IVUS images for
SVM Classifier with Modified Radial Basis Function Kernel accurate morph metric analysis of coronary plaques and
(MRBFK). accordingly the assessment of the atherosclerotic lesion
length [7]-[9].
Key words: Atherosclerosis, Coronary Heart Artery,
Classification, Feature Extraction, Lumen diameter, Shape Cardiac disease research is the most important areas in the
Parameter. field of medicine and also that contains the advantages of
several existing technologies [10]. Predominantly coronary
1. INTRODUCTION heart artery disease and atherosclerosis are worldwide fatal
diseases for both men and women among all ethnic groups.
Health care systems are based on the progressive information American Heart Association says the ratio of death rate is mainly
technology to electronically accumulate process, scrutinize, increased based on cardiac diseases. Atherosclerosis involves a
build-up on the inner side of artery walls [11]. Heart arteries
are supplies the blood to the heart. Normal artery contains the
3877
S. Sridevi et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(3), May – June 2020, 3877 – 3886

normal blood flow through the artery and up normal arteries concentrates only on Masking and segmentation process. The
contains the up normal blood flow through the artery which is above limitation can be overcome by using the following
usually leading to a heart attack and cardiac arrest to the proposed method.
blood supply. There is overwhelming evidence that high
blood cholesterol increases the risk of developing The paper is organized as follows: Literature survey is
atherosclerosis [12]-[15]. described in section 2. Proposed methodology are detailed in
section 3 followed by the statistical analysis and performance
This system is also able to perform risk classification of evaluation. In section 4 described the results and discussion
patients based on the semantically structured information of and comparison results are in section 4.1. Finally, the
patient data. And also contains the detection of luminal and conclusion and future enhancement are given in section 5.
medial–adventitial boundaries, based on the results of
neighbor pixel analysis, performed by means of a Kernalized 3. MATERIALS AND METHODS
function, and on a smoothing step using Radial Basis The flow chart of the proposed method for classification of
Function (RBF) approximation [16]. coronary heart artery disease using IVUS image is shown in
Figure 1.
2. LITERATURE SURVEY

Ji Yang, Mehdi Faraji [17] described the FCN based deep

architecture for automatic segmentation of IVUS frames and
also reconstructing the human arteries. Deep architecture
using a smaller number of training images and achieves the
higher degree of generalization ability and also strengthen the
augmentation process. Training images are evaluated with
commonly used parameters like Jaccard measure and
Hausdorff distance. The above said method not evaluated the
Figure 1: Flow chart of proposed method
other standard parameter like specificity, precision accuracy.
Lucas Lo Vercio, Jose Ignacio Orlando, Mariana del Fresno, 3.1 Image Acquisition
and Ignacio Larrabide [18] presented a automatic
Image acquisition is the first stage in fundamentals of digital
segmentation of luminal boundary by means different image
image processing. After acquired an image, various image
features. Image filters, edge detectors and textual descriptors
processing methods are applied on images for further
were taken for the assessment. The assessment is based on
processing. Here medical heart image data set is taken for
classification techniques. SVM classifier used to training the
processing to find out the disease of coronary heart artery [21],
feature set and AUC and Jaccard index measurement.
[22].
VaishaliNaik, R.S.Gamad and P.P Bansod [19] explained the
3.2 Image Preprocessing
segmentation of the common carotid artery and determination
of the Intima-media thickness based on the commonly used This section includes the following important steps are
methods are dynamic programming techniques, Hough A. Removal of Shadow Reduction/Blood noise reduction.
transform, Nakagami mixture modeling, active contour and B. Cartesian to Polar (C2P) Coordinate conversion.
edge detection and tracking method. And then performance is
validated by different parameter metrics. A. Removal of shadow reduction/Blood noise
Huang Yi; Yan, Wenjun; Xia, and Menghua [20] explained a reduction
special mask based on the geometric property of media and an
Shadow/blood noise areas usually do not represent any useful
assumption about the intensity distribution. For media
information in grey scale IVUS image. Therefore, this method
segmentation, the guided multi scale normalized cut extract
the Region of Interest (ROI). Then a target binarization has two types of thresholds namely High threshold and low
processing can extract the rough lumen region of ROI. threshold. Thresholding techniques are used to detect the low
Geometrical and binarization methods are used in the ROI intensity regions following the high intensity regions, because
extraction. Based on the observation the above system
3878
S. Sridevi et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(3), May – June 2020, 3877 – 3886

the very high intensity regions are might belong to the plaque
regions of IVUS image.

If this system denote intensity of the pixel I (i,j) within the input
image and the pixel x is the shadow region and it can be
defined as follows:

x= (i,j) € ∑
Figure 3: Transformation of IVUS image regions
If intensity of pixel < low threshold
and I(i, j±1) ɛ ∑ (1) Iimage (rradial, θtangential) (3)
If intensity of pixel < low threshold
It is also facilitates the number of detection steps for
and I(i, j±1) > high threshold (2)
Ultrasound medical images such as contour initialization and
refining of the obtained contour in IVUS image Regions and
the algorithm 1 is used to describe the angle decomposition for
coordinate conversion.

3.3 Iterative level Energy Computation Method (IECM)

Figure 2: Separation of border detection from the background
IECM consists of Region-based geometric contour models are
Figure 2 illustrates two sample images from the dataset, here
used to detect the connected pixels that employ the region
the yellow line indicates the shadow region and red line
information from the IVUS images. All possible integer
indicates the border detection. The threshold values are
threshold values are binarize and analyze the local edge base
selected based on the mean value of the intensities in the
results are obtained by IECM. IECM contains the two
shadow regions like high threshold and low threshold
important steps of region based geometric contours which are
correspondingly. The resulting threshold values are initially
extracted from the input image.
normalized into statistical unity like zero mean and standard
deviation for generalization purpose. Algorithm 2 explains the first step includes the region that
evolves from brighter surfaces to darker boundaries is
B. Cartesian to Polar (C2P) Coordinate conversion
identified by the variable C1-. The superscript ‘-‘symbol
Coordinate conversion is the important for describing the local indicates that the brightness point of an image values. The
image regions. So, the input image is convert from Cartesian to values decreased from the surface of the image regions towards
polar domain format which is shown in figure 3, where rows the boundaries of an IVUS image. The second step consists of
and columns are corresponding to the angle and distance from region that evolves from darker surfaces with brighter
the center of the catheter device space which is denoted as boundaries and is denoted by C2+. The above said region is
I(r,θ)as shown in eq 3. used to visualize the characteristics of IVUS image layers
(Lumen and media). Therefore, IECM need to extract the C2+
regions are used to obtain the information about the lumen and
Algorithm 1 Angle Decomposition of IVUS Image regions
media because both the regions evolve from dark surfaces with
Step (1) Input : IVUS image size of 512 x 512
brighter boundaries.
pixels
Step (2) img : bidimensional image. Algorithm 2 Working Procedure of IECM
Step (3) Radius : radius length (# of pixels to be Input image with initial contour
considered).
Computes C1, C2, average prototypes of regions
Step (4) Angle : # of angles to be considered for
for inside and outside of the contour
decomposition.
New degree of membership value is calculated
Output: pcimg: polar coordinate image.
Difference Δf between new and old energy is calculated
Calculate total energy F
Output: Get a segmented image

3879
S. Sridevi et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(3), May – June 2020, 3877 – 3886

3.4 Feature Extraction Euclidean two dimensional space, then the distance (d) vector
from p to q, or form q to p is defined as
In feature extraction stage, most important key points are
captured and extracted from IVUS images. In any
segmentation approach considerable number of normal d(p,q) = d(q,p) = (q1-p1)2+(q2-p2)2+…+(qn-pn)2 (8)
objects is recognized as pathological, which results in
reduced efficiency of the detection system. However, one 3.4.3 Nearest neighbor approach
of the most important problems of IVUS imaging system is
SIFT uses the nearest neighbor approach and it is mainly used
selection of standard feature set and the classification for classification method, which should be the first choice
method, which is used to extract Region of Interest (ROI) for a feature extraction study, where there is little or no more
while minimizing the false positive findings. Feature prior knowledge about the distribution of pixels in an image
extraction technique and its corresponding features are shown in the following Figure 4 – 6 and algorithm 3.
reported in Table 1.
Algorithm 3 nearest neighbor approach
Ta b l e 1 : Feature extraction techniques with corresponding
feature NN (Image, Candidate points) {
Step (1) Go through each candidate points in an image,
Feature extraction techniques Step (2) Calculate the "distance" from that image to
specific sample candidate points.
Shape: X-coordinate maximum length of the plaque image, Step (3) Out of those candidate points values pick the "k"
y-coordinate maximum length of the plaque image, area of Ones that is most close to candidate
ROI, Perimeter of ROI. Eccentricity of Lumen. points, maybe in a premade array of "k" }.
SIFT: Finding Harris corners, Calculating characteristic
scale Calculating main orientation, Calculating descriptors, Output: Get a Premade
Plotting/rotating matches.
In polar coordinates of the point p are (radial 1 characteristic
LBP: Relation between pixels and its neighbors.
(r1), tangential 1charactrestic (θ1)) and those of q are (radial
3.4.1 Shape Features 2 characteristic (r2), tangential 2 characteristic (θ2)) than the
distance between the points are defined as follows.
Shape features are mainly used to extract the following features
to identifying the exact plaque regions in a coronary artery. The Polar point = (R2+r2-2r1r2COS(θ1-θ2))2 (9)
features are like x and y coordinates maximum length, Region of
Figure 7 explains the Image feature points and also detected
Interest (ROI), perimeter of ROI and Eccentricity value
by using the Gaussian difference function, thus providing the
calculation.
similarity invariance between corresponding points of an
image and also creating the feature descriptor by using the
The above said features are providing essential information to
common geometric transformation.
classify the heart disease dataset according to the mathematical
calculation shown in following equations.
Scale invariant interest point detector consists of four
parameters like initial Gaussian blur, steps per scale octave,
D=2 (4) minimum image size and maximum image size. Local image
CSF = (π.D/observed perimeter) 2
(5) interest points are matched by using local image descriptor. It
SCSA = ((CSAm – CSAI)/CSAm) x 100 contains the following similar descriptors are feature
(6) descriptor orientation bins, feature descriptor size, closest or
next closest ratio. Local image feature descriptions are
LLMAXD-LLMIND/100 (7)
filtered by using the geometric filter, which are maximum
3.4.2 SIFT Features alignment error, minimal inliers ratio, and expected
In this approach introduces the Nearest Neighbor approach with transformation.
high probability Euclidean metric based algorithm for matching
the feature points in an image. In Cartesian coordinates, if p =
(p1, p2,…pn) and q = (q1, q2,… qn) are two points in

3880
S. Sridevi et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(3), May – June 2020, 3877 – 3886

3.4.4 LBP Features

Local Binary Pattern (LBP) is a type of feature used in

classification. Local feature encode for texture information of
IVUS images, Table 2 explains the feature vector information
which can use for IVUS image classification to obtaining the
region information. The function portioning input image
information for non-overlapping cells. The working
procedure of LBP method is shown in algorithm 4.

Algorithm 4 Generation of LBP Feature Vector

Step (1) The number of cell is calculated by using the
Numcells = prod (floor (size (I) cellsize)) Figure 5: Finding nearest neighbor approach with specified
Step (2) Feature length calculation using radius
LBP feature N=Numcells x B
LBP Features = size(Lyx) [1,B]
Output: Get a Texture information

Ta b l e 2 : LBP features for obtaining textual information

about IVUS images
Input Algorithm Histogram Output
Arguments Parameter Parameter Arguments
Number of Features
neighbors Normalizatio
select n Call for
neighbors Interpolation number
Input Figure 6: Combining number of neighbors and radius
used to of cells
image
Upright compute Call for
(Rotation pixels. LBP
Invariance Feature
Flag) Length

Figure 4: Selection of Nearest Neighbor

Figure 7: Enhancement of SIFT descriptor

3.5 Classification

In classification SVM uses the Modified Radial Basis

Function Kernel (MRBFK). To avoid the misclassification
3881
S. Sridevi et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(3), May – June 2020, 3877 – 3886

error, the training classifier finds out the optimal hyper plane
and it is used to map the input vector elements into higher
Algorithm 5 SVM – MRBFK Parameters cross validation
dimensional space where a hyper plane can separate the data
procedure
into different classes (Actual and Predicted classes).
Classification begins......
The Radial basis function is mainly handling the non-linear (1) ClassMidpointNormalize(Normalize);
relations with classes and their feature vectors and then (2) load and prepare the dataset
(3) Dataset for grid search
normalization was applied into feature vectors [0, 1] before
(4) x = IVUS.data
applying the classifier. Standard deviation and the radial (5) y = IVUS.target
basis kernel were selected based on the feature selection. The (6) Dataset for decision function visualization: features
kernel function considering two parameters namely k (xi,x) in x and subsamples to keep only 2 classes.
and γi. The k is the kernel parameter and γi is the error term. (7) x_2d = X[:,:2]
Testified results were obtained in order to set up the kernel (8) Scale the data for SVM training
(9) Training set obtains the transformation from one
parameter is shown in Figure 8.
coordinate to another.
MRBFK represents the value of inner product of two training x_2d=scalar.fit.transformation(x_2d) %for an
points in feature space. Kernel function uses the two steps initial search set the gamma range.
which is mapping and do inner product in the feature space. (10) Param _grid = dict(gamma = gamm_range, C =
The Modified Radial Basis Function Kernel represented as C_range)
feature vector in some input space and used to maps the (11) validate accuracy of proposed method MRBFK
samples namely αi and xi, is defined as with the variable (gamma and C)
Proposed model of the SVM with Modified Radial Basis
Function Kernel
Step 0: Set the control parameters C,σ, k
Step 1: Train SVM with a primary MRBFK defined by Eq
Therefore (10) 7
Get αi, b and the set of support vectors and determine the
decision boundary by f(x) = ∑γiαiK(x, xi) + b iєsv
3.5.1 SVM Classifier with MRBFK Step 2: given f(x), train the modified RBFK SVM by Eq 8
For the same data and get the new decision boundary ḟ(x)
Modified Radial Basis Function Kernel is used to performing Step 3: based on ḟ(x) and newly observed state is
the models are easier to finding the best parameters place on obtained.
the IVUS image boundaries of the grid, and also kernel Output: Calculate the prediction and store the result.
parameter can be extended up to subsequent search.
Algorithm 5 consists of two steps, Step 1 initially, is used to
simplified classification problem involving the input features
(Shape, SIFT and LBP) and 2 possible target classes (actual
and predicted class). Finally, decision functions are
visualized for variety of parameter values. Step 2 consists of
mapping the classifiers of cross validation accuracy as a
function C and gamma. This paper found the effectiveness of
the SVM equipped with MRBFK to improve the precision and
studied the specificity with respect to the parameter k which
controls the degree of the conformal transform is defined as
follows

f(x) = ∑_i^Nαiγik (xi,f(x) + β (11)

Figure 8: MRBFK Creation using feature vectors based on training
f(x) = ∑_i^Nαiγiexp√(- (((|(|x-x_i | )| ))⁄2σ)) + β (12) and testing set. MRBFK, Modified Radial Basis Function Kernel

3882
S. Sridevi et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(3), May – June 2020, 3877 – 3886

3.5.2 Statistical Analysis Table 3: Pseudo code for performance metrics (TP, FN, TN, and TP)
True False True False
In this section the specificity, sensitivity and accuracy of Positive Negative Negative Positive
classifier validation studies are calculated. In addition the Procedure: Procedure: Procedure: Procedure:
Kappa Statistic, Area under the Curve (AUC) and
Calculate calculate Calculate Calculate
Confidence Interval (CI) level are also computed from the the True the False the True the False
above said parameters for quantifying the degree of Positive as Negative Negative Positive as
agreement defined as follows. follows. as follows: as follows: follows:
True Flase True Flase
Kappa Value= ((observed – expected) Proportion / 1 – Positive Negative Negative Positive
expected proportion). (13) Rate= TP / Rate = Rate= TN / Rate = FP /
(TP + FP). FN / (FN + (TN + (FP + TN)
Area Under the Curve y = f(x) between x =a & x = b, Integrate FN).
I – Read TP). I – Read
y=f(x) between the limits a and b. (14)
image, T – I – Read I - Read image, T –
Ground image, T – image, T – Ground
3.5.3 Performance Evaluation truth Ground Ground truth
image. truth truth image.
TP=1 image. image. FP = 0
In the proposed method used to evaluate the performance
For i : m FN = 0 TN=1 For i:m
through statistical analysis. For this, True Positive (TP), False
Positive (FP), False Negative (FN) and True Negative (TN) step of 1 For i:m For i:m step of 1
For j : n step of 1 step of 1 For j: n
and then Sensitivity, Specificity, Precision and Accuracy are
calculated and shown in Table 3. step of 1 For j: n For j:n step step of 1
If I(i, j) = 1 step of 1 of 1 If (i, j) = 1
CI (0.86%) = X ±1.96 x Xx (1 – X)/ N (15)
and T (I, j) If (i, j) = 1 If I(i, j) =0 and T(i, j)
= 1 then and T(i, j) and T(i, j) = 0 then
4. RESULTS AND DISCUSSION
TP = TP = 0 then = 0 then FP = FP +
+1, print FN = FN + TN = TN + 1, print
In this section, the performance of the proposed method is True 1, print 1, print Flase
presented. First, the performance of shadow detection Positive False True Positive
procedure investigated and corresponding segmented IVUS (TP) Negative Negative (FP).
images are analyzed by the proposed method. A sensitivity of (FN). (TN).
93.6 and specificity of 96.6 is obtained by comparing
manually and automatically detected regions. After shadow Condition: Condition: Condition: Condition:
detection the Shape, Scale Invariant Feature Transform TP Rate= 1 FN Rate = TN Rate = FP Rate =
features and Local Binary Pattern features were detected from when FP = 0 when 1when 0 when FP
the remaining portion of the plaque area. 0. FN =0. FN = 0. =0.
(No false (No false (No false (No false
positives) negatives) negatives) positives)

Sensitivity, Specificity, Precision and Accuracy parameters are

calculated in order to assess the performance of the proposed
method. Figure 9 shows the calculation of three statistical
parameters which are sensitivity, specificity and precision.
Figure 10 shows accuracy is calculated based on the
Parameters like true positive, false positive, false negative and
true negative.

Figure 11 shows the Confusion matrix of MRBF Kernel

creation. Confusion matrix is a special kind of contingency
table which allows the visualization of the performance of a

3883
 S. Sridevi et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(3), May – June 2020, 3877 – 3886

method. It is a specific table layout, with two dimensions 4.1 Comparison Results
(“actual” and “Predicted”), and identical sets of “classes” in The Existing and proposed method and its corresponding
both dimensions. Confusion matrix shows the performance of metric based analysis are reported in the table 4. According to
the proposed Modified Radial Basis Function Kernel the feature extraction, results based classification results are
(MRBFK) which are effectively used in higher dimensional validated and also the proposed method performance are
space to detect the misclassification errors in training and reported and compared with the existing methods.
testing samples.
Existing Method 1 (SVM) and its corresponding metrics
The Classification accuracy is reported in table 4. Classifier (Sensitivity, Specificity, Precision and Accuracy) is
that gives nearly similar accuracies which shows the reliability evaluated. Overall accuracy is 74%.
of the proposed image based validation method.
Existing Method 2 (ECOC) and its corresponding metrics
(Sensitivity, Specificity, Precision and Accuracy) is
evaluated. Overall accuracy is 84 %.

Proposed MRBFK and its corresponding metrics (Sensitivity,

Specificity, Precision and Accuracy) are evaluated. Overall
accuracy is 88%. All the above results have been tabulated in
Table 4 and have been shown graphically in Figure 12.
Figure 9: Graphical representation chart for evaluation of
performance of classification parameter like Sensitivity, Table 4: Comparison results
Specificity and Precision

Figure 10: Graphical representation chart for evaluation of

performance of classification parameter like True positive,
False positive, True Negative and False Negative

Figure 12: Graph representation of comparison result

5. CONCLUSION AND FUTURE ENHANCEMENT

In this paper, an effective method for the challenging

problem of Coronary Heart Artery Disease classification
Figure 11: Graphical representation chart for confusion was presented. Several combining features were produced
matrix of SVM classifier with MRBFK based on the
in this work.
metrics like True Positive, False Positive, False
Negative, True Negative and Accuracy The SVM with MRBFK is used to describe the reliability
and projection into this infinite higher dimensional space
where the data becomes linearly separable. Then, the Radial

3884
S. Sridevi et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(3), May – June 2020, 3877 – 3886

Basis Function Kernel is used when the boundaries are 8. Hassen Lazrag, Kamel Aloui, and Med Saber Naceur.
hypothesized to be curve-shaped. Another contribution of Automatic Segmentation of Lumen in Intravascular
this paper investigates thoroughly on new data sets and also Ultrasound images using Fuzzy Clustering and
asses the reliability of the reported results. Active Contours, International conference on control,
engineering & Information Technology (CEIT ’13)
The results supporting the positive effects of combining the
Proceedings Engineering & Technology- Vol.1, 2013.
three feature extraction methods. And then, generates the 9. Luo Z, Wang Y, and Wang W, Estimating coronary
multi – resolution characteristics of more detailed images artery lumen areawith optimization-based contour
performs better. Furthermore, the results obtained might be detection, IEEE Trans Med Imaging, 22(April):546–64,
the (88%) highest achievable accuracy, using the 2013.
combination of image based feature extraction method. https://doi.org/10.1109/TMI.2003.811048
In future research, there is a need and huge scope to improve 10. Sridevi S, Sundaresan M. Hybrid Feature Extraction
the Dice coefficient, Recall, and F-measure in order to Techniques For Accuracy Improvement In IVUS
Image Classification, International Journal of
perform the classification of IVUS images.
Scientific & Technology Research, Volume 9, Issue 04,
ISSN-2277-8616, 2020.
REFERENCES 11. Sun Z, Liu C,A. parallel method for segmenting
1. Arash taki,Holger hetterich and et al. A New approach intravascular ultrasound image sequence. Appl Mech
for Improving Coronary Plaque Component Mater, 130:2051–5, 2011.
Analysis Based on Intravascular Ultrasound Images, 12. Syaiful Anam, Eiji Uchino and Noriaki suetake. Hybrid
Elsevier, Ultrasound in Med. & Biol., Vol. 36, No. 8, Boundary Detection Method for Image with
pp. 1245–1258, 2010. Application to Coronary Plaque, International
2. Archana K.V.,.Vanithamani R. Review on Journal of Digital Information and Wireless
Intravascular Ultrasound Image Detection, Communications (IJDIWC), Vol. 4, 2014.
International Journal of Advanced Research in 13. Topol E.J. Textbook of Cardiovascular Medicine,
Science, Engineering and Technology, Vol.4, Issue 7, Lippincott Williams & Wilkins, 3rd edition, 2007
2017. 14. Xuan Liang, Michalis Xenos, Yared Alemu, Suraj H and
3. Ashok kumar L., and P. Rajendran. Intima-media et al. Biomechanical Factors in Coronary vulnerable
thickness segmentation using weighted graph based Plaque Risk of Rupture: Intravascular
active contour technique, Biomedical Research, 1-1, Ultarsound-based Patient- Specific fluid – structure
2017" interaction studies, Pathophysiology and natural
4. Christodoulou C I, Pattichis C.S. Image Retrieval and history, Vol 24, No2, 2013.
Classification of Carotid Plaque Ultrasound Images 15. Selvathi D, Emimal N. Performance analysis of
The Open Cardiovascular Imaging Journal, 2, 18-28, contourlet Features with SVM classifier for the
1876-5386, 2010. characterization of Atheromatous plaque in
https://doi.org/10.2174/1876538601002010018 Intravascular Ultrasound images, Journal of
5. Daisuke Nagatomo, Daizaburo Yanagi, and Takeshi Engineering Research and Applications, ISSN:
Serikawa. The Role of Intravascular Ultarsound 2248-9622, vol.4, Issue 3(Version 4), pp.35-40, 2014.
scan and thin-sliced coronary computed tomography 16. David Molony, Habib Samady. TCT-342 Deep IVUS:
angiography in diagnosing aortic dissection causing A Machine Learning Platform for Fully Automatic
acute myocardial infarction, International Journal of IVUS Segmentation and Phenotyping, Journal of the
Case Reports and Images, vol.5 no 4, 2014. American College of Cardiology, Volume 74, Issue 13
6. Escalera S, Pujol O, and PetiaRadeva. ECOCs library, Supplement, DOI: 10.1016/j.jacc.2019.08.424, 2019.
J. Mach. Learn. Res. 11 661–664, 2010. 17. Ji Yang , Mehdi Faraji. Robust Segmentation of
7. Ga Young kim, Ju Huwan Lee. A Novel Intensity Arterial Walls in Intravascular Ultrasound Images
based Multi level Classification approach for Using Dual Path U-Net, Ultrasonics National Library
Coranry Plaque Characterization in Intravascular of Medicine, DOI: 10.1016/j.ultras.03.014, 2019.
Ultrasound Images, Biomed Engineering online, 18. Lucas Lo Vercio, Jose Ignacio Orlando, Mariana del
DOI:10.1786/S12938-018-0586-1, 2018. Fresno, and Ignacio Larrabide. Assessment of image

3885
S. Sridevi et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(3), May – June 2020, 3877 – 3886

features for vessel wall segmentation in

Intravascular Ultrasound images, Article in
International Journal of Computer Assisted Radiology
and Surgery 11(8), DOI: 10.1007/s11548-015-1345-4,
2016.
19. VaishaliNaik, R.S.Gamad and P.P Bansod. The Carotid
Intima media Thickness Measurement of
Ultrasound images”, Proceedings of the 12th
INDIACom, INDIACom-2018; IEEE Conference ID:
42835, ISBN 978-93-80544-28-1, 2018
20. Huang Yi; Yan, Wenjun; and Xia, Menghua.
Segmentation of Media and Lumen in Intravascular
Ultrasound Image using Guided Multiscale
Normalized Cut, Journal of Medical Imaging and
Health Informatics, Volume 9, Number 7, 361-364,
2019.
21. B. Jabber.at.al, "A Novel Sampling Approach for
Balancing The Data and Providing Health Care
Management System by Government", International
Journal of Advanced Trends in Computer Science and
Engineering, Vol 8, No. 6,November -December 2019,
2753 – 2761.
https://doi.org/10.30534/ijatcse/2019/12862019
22. Mani V, GokulPrasath R, Jegatha S, et al, “Hospital
Recommendation System using Machine Learning”,
International Journal of Advanced Trends in Computer
Science and Engineering, Volume 9, No. 2, March-April
2020, https://doi.org/10.30534/ijatcse/2020/64922020.

3886