International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
Retinal Vessel Segmentation Using an
Entropy-Based Optimization Algorithm
Sukhpreet Kaur, IKGPTU, Kapurthala, India
Kulwinder Singh Mann, Guru Nanak Dev Engineering College, Ludhiana, India
ABSTRACT
This article presents an algorithm for the segmentation of retinal blood vessels for the detection of
diabetic retinopathy eye diseases. This disease occurs in patients with untreated diabetes for a long time.
Since this disease is related to the retina, it can eventually lead to vision impairment. The proposed
algorithm is a supervised learning method of blood vessels segmentation in which the classification
system is trained with the features that are extracted from the images. The proposed system is
implemented on the images of DRIVE, STARE and CHASE_DB1 databases. The segmentation is
done by forming clusters with the features of patterns. The features were extracted using independent
component analysis and the classification is performed by support vector machines (SVM). The
results of the parameters are grouped by accuracy, sensitivity, specificity, positive predictive value,
false positive rate and are compared with particle swarm optimization (PSO), the firefly optimization
algorithm (FA) and the lion optimization algorithm (LOA).
KEywORdS
Diabetic Retinopathy, Feature Extraction, Optimization, Retinal Vessels
INTROdUCTION
The structure of blood vessels in retina helps in detection of number of eye diseases which includes
arteriosclerosis, diabetes, retinal vein occlusion, retinal artery occlusion, hypertension, cataract,
glaucoma and most importantly diabetic retinopathy. These all diseases can be detected by monitoring
the changes in the structure of an eye. A human eye consists of iris, lens, blood vessels, pupil, retina
etc. Eye helps in sensing and visualizing different objects. All the different parts of an eye help in
visualizing in one or another way. Each and every part can lead to different disease if affected by
diabetes. The patients having prolonged and untreated diabetes suffered from eye disease named as
diabetic retinopathy (DR). DR is the leading cause of blindness as the retina of the eye is directly
affected by this disease. According to the latest figures issued by World Health Organization (WHO),
the patients suffering from diabetes will reach to 300 million by 2025 (Zimmet, 2016). Currently, the
number of diabetics are 69.2 million from which 7 million people suffer from vision loss (Joshi, 2016).
Diabetes can affect any body part like eyes, kidneys, the liver, the heart, and bones. Eye became
the significant part of the human organ system needs special care. The impact of untreated blindness
DOI: 10.4018/IJHISI.2020040105
This article, originally published under IGI Global’s copyright on December 20, 2019 will proceed with publication as an Open Access
article
starting
on January
14, 2021
in the or
gold
Open Access
journal,
International
Healthcare
Information
and
InformatCopyright
© 2020,
IGI Global.
Copying
distributing
in print
or electronic
formsJournal
withoutofwritten
permission
of IGISystems
Global is
prohibited.
ics (converted to gold Open Access January 1, 2021), and will be distributed under the terms of the Creative Commons Attribution License
(http://creativecommons.org/licenses/by/4.0/) which permits unrestricted use, distribution, and production in any medium, provided the
author of the original work and original publication source are properly credited.
61
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
is shown on blood vessels, nerves and the vision of the patient. DR is caused when the blood vessels
of the human retina are damaged, and they leaked blood and lipids. The symptoms of DR is same
as that of changes that occur in eyes due to age, so there is a need of fine procedures which can
differentiate between DR and age related eye degradation.
DR can be detected easily by segmenting retina and its blood vessels. The segmented images
of the retina help in studying the blood circulation of human eye at the micro level. As it is the part
of central nervous system, so it is easy for researchers as well as for the ophthalmologists to study
the retina for different pathologies (Fraz, 2012). It is highly sensitive to light and consists of optic
disc, blood vessels and macula. The various pathologies in the retina of an eye can be detected by
monitoring the variations in the various components of retina. The blood vessels can be easily visible
to the human eye. So, the pathologies in blood vessels can be checked easily by the clinicians. All
the different eye diseases that occur due to pathologies in retina can be detected by segmentation
of retina. The retinal images are captured using special camera named as Fundus Camera as well as
by using ophthalmoscopes. Fundus camera is a camera of high resolution especially used for retinal
imaging. Other techniques used for acquiring retinal images include laser screening, optics screening
and angiography. Fundus imaging is prominently used for retinal imaging by dilating the pupil of
retina using some eye drops. Then the fundus of the image which is the region opposite to lens of
eye and includes optic disc and macula is focused for imaging.
The various diseases in eye cause different types of changes in the vasculature of human retina.
The various disorders of an eye can be checked by studying the segmentation of retina and its various
parts. The clinicians also study the changes in the retinal vasculature for evaluating the severity of eye
diseases and to decide whether the disease can be curable or not. The various changes in the retinal
eye can be categorized into neovascularization, collateralization and origination of retinal vascular
shunt (Paul, 1974). If the new blood vessels are originated in either retina or in the area adjacent to
it, then it leads to neovascularization. In this the blood vessels grows in irregular fashion generally
near larger arteries and veins in any direction. They are appeared in the areas where there are no
blood vessels present. Collateralization relates to the growth of blood vessels in between the existing
blood vessels by joining new arteries and veins to the existing new arteries and veins respectively.
The blood flow in the vessels is hampered if there is cross connection between them. The last case of
formation of shunt occurred when the blood flows without using capillary bed at a very high speed.
DR is disorder of human eye which is caused by untreated diabetes. In this the blood vessels are
damaged and they leak blood and in some cases they lead to growth of new blood vessels. Due to
which, the vision deteriorates and it leads to blindness. DR leads to formation of microaneurysms,
exudates, hemorrhages, cotton wool spots, and lesions. Microaneurysms are small red dots which are
formed when the walls of capillary blood cells are weakened. When the weakened blood cells leak,
they become hemorrhages which are flame-shaped. After that, when the proteins and lipids from
the blood are leaked, then they lead to formation of exudates. Hard exudates are of yellow or white
color in eye retina. When the severity of DR advances, the blood vessels get obstructed and leads to
soft exudates or cotton wool spots and they are white in color.
DR is classified broadly into two stages: Proliferative DR (PDR) and Non-Proliferative DR
(NPDR). NPDR is the initial stage of DR in which the damage of retinal blood vessels has just started.
The three stages of NPDR are mild, moderate and severe (You, 2011). The mild stage of NPDR is
the initial stage of NPDR and requires no treatment but the progression of the disease needs to be
monitored strongly by the clinicians. In mild NPDR only microaneurysms occur or in some other
cases hemorrhages can also occur. In moderate NPDR, cotton wools spots start appearing due to
blockage of blood vessels that nourish the retina. In the case of severe NPDR, the growth of new
blood vessels starts in an irregular fashion in the eye retina. Finally, vision loss or blindness occurs
due to formation of new blood vessels and due to weakness of existing blood vessels in the retina.
DR can be detected by the segmentation of blood vessels of the retina. The variations in width,
length, branching angle, vascular pattern, and tortuosity of the blood vessels can be helpful in detecting
62
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
various eye diseases. Manual segmentation of the human retina is a tedious task and requires expertise.
So, various computer-aided techniques have come into existence which helped ophthalmologists
to detect eye diseases to help their patients from vision impairment. Although there are number of
techniques, but there is always a way of improving the existing methods.
BACKGROUNd
The existing techniques of retinal blood vessels segmentation was categorized into seven major
categories named: a) machine learning or pattern classification techniques; b) matched filtering; c)
mathematical morphological techniques; d) vessel tracing; e) multiscale approaches; f) model based
approaches; and g) hardware-based approaches. The large number of existing techniques is under
the category of pattern classification techniques which categorizes the extracted features either into
vessels or non-vessels. The pattern classification techniques are further divided into supervised and
unsupervised techniques. In the case of unsupervised techniques, the classification of vessels and
non-vessels is done based on extracted patterns.
Supervised Methods of Pattern Classification
In case of supervised techniques, the training data or the ground truth images are available for
training the system. The ground truth images are generally segmented by the experts or clinicians.
The classification of blood vessels is done by comparing the features of extracted patterns and the
patterns of ground truth images. Li, Zhenshen, Chao et al. (2017) proposed a supervised technique
for blood vessel segmentation based on features given by length, width and intensity of the input
images. Since these are the local descriptors of the image, so they helped in maintaining the edge
information locally. Mustafa, Haniza and Wahida (2017) proposed a technique for the detection of
diabetic retinopathy and glaucoma based on 7-dimensional feature vector that was hybrid of gray
levels and moment invariants features. The classification of vessels and non-vessels was finally done
by using decision trees. Shah, Tong, Ibrahima et al. (2017) proposed a technique that can find the
abnormalities of human retina using regional and Hessian matrix descriptors. The total number of 24
features was extracted for the pattern recognition process and the classification was performed using
linear minimum squared error method and this helped in achieving the accuracy of 93% approximately.
GeethaRamani et al. (2016) proposed a supervised method for the segmentation of blood vessels
based on both image processing and data mining. The supervised learning was performed by the
combination of k-means clustering. The final segmented image was formed by using decision tree
classification and image post-processing by mathematical morphology and connected component
analysis. Rahim et al. proposed a technique for the detection of microaneuysms using features given
by area, length, mean, perimeter, major and minor area length, standard deviation. The classification
process was performed by using decision trees, k-nearest neighbors, support vector machine, RBF
Kernal SVM. This method was highly helpful in detection of DR as microaneuysms exists during
the initial stages of DR. Aslani and Sarnel (2016) compute the 17-dimensional feature vector for
the segmentation of blood vessels. The classification was done using Random Forest (RF) classifier
for dealing with both with homogeneous and heterogeneous data. RF classifier is based on decision
trees and it was implemented in the proposed algorithm using 150 decision tree with a depth of
15 for each branch. Hatanaka, Samo, Tajima et al. (2016) proposed a supervised technique for the
segmentation using autocorrelation method on shift invariant local features on neighbors of pixels.
The classification process was done using neural networks of two types in which first neural network
worked on 105 pixels while the second neural network worked on output of first neural network,
filtering and transformation process. The advantage of this method is that it can segment the images
with low contrast.
Saha, Naskar, and Chatterji (2016) proposed a technique for detection of DR using wavelet transform
and neural network. The wavelet-based analyzer was used to analyze the segmented images with ground
63
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
truth images. The feed forward neural network was used for classification of vessels and non-vessels
in the segmented images. Rajput, Manza, Patwari et al. (2015) proposed a method to detect NPDR in
which features were extracted using Haar and Symlet wavelets and they will extract the features for each
symptom of DR. This process of feature extraction helps in classifying the input image into different
stages of NPDR that is mild, moderate or severe by extracting patterns for each symptom separately.
The classification was finally done by k-means clustering and statistical means. Tang, Lin, Yang et al.
(2015) designed a feature vector of 94 dimensions for pattern recognition. The features include Gabor
responses at different scales and dimensions. The classification process was performed using support
vector machines (SVM). Franklin et al. gave a supervised technique using neural network for the
segmentation of retinal blood vessels. The image was preprocessed using background normalization and
the system was then trained using feed-forward perceptron neural network and the classification was done
using back-propagation algorithm. Franklin and Rajan (2014) used artificial neural network for training.
The extracted features include Gabor responses and moment invariant based features. The input image
was preprocessed using different filtering techniques and the system was trained and classification was
done. The image was also post processed after connecting different isolated points. Wang, Yin, Cao
et al. (2014) proposed a technique using ensemble learning for the segmentation process. In this, two
classifiers were used, random forest (RF) classifier and convolutional neural network (CNN) which
act as a feature classifier and feature extractor, respectively. CNN works in two different sublayers as
convolutional sublayer and subsampling sublayer for extraction of features. RF classifies the vessels using
the majority voting process and using winner –takes-all decision strategy. Roychowdhury, Koozekanani
and Parhi (2017) used a Gaussian Mixture Model (GMM) for the classification of prominent features.
The features were extracted using gradients of first and second-order derivatives. The key feature of
this proposed algorithm is that the input image is converted into binary image for extraction of features.
Marin, Awuino, Gegundez et al. (2011) designed a 7-dimensional feature vector for the classification of
input images. The extracted features include gray level and moment invariant based features. The input
image was preprocessed using central light reflex removal, background homogenization methods. The
training was done using neural networks by calculating probability map of each pixel. The classification
was performed using thresholding process. Finally, the gaps were filled in post-processing phase using
artifact filling process. Peng et al. proposed an algorithm that can segment both thin as well as wide
vessels by using radial projection. The thin and narrow vessels were segmented using radial projections
and the wide vessels were segmented using steering wavelet and semi supervised learning. Lupascu,
Tegolo and Trucco (2010) used a feature-based Ada Boost classifier for the segmentation of retinal blood
vessels. The constructed feature vector is of 41 dimensions including features of intensity, spatial and
geometrical features. Xu and Luo (2010) proposed a segmentation process using wavelets and curvelets
for both thin and thick vessels. The thin vessels were classified further using Hessian matrix and thick
vessels by support vector machines. Osareh and Shadgar (2009) proposed a method for segmentation
for colored images of retina using principal component analysis, Gaussian model and support vector
machines. Anzalone, Bizzari, Parodi et al. (2008) proposed an algorithm consisting of two blocks in
which first block was used for enhancement of images of blood vessels and second block was able to do
image binarization, image cleaning after evaluating the optimized values of measures of performance
(MOPs). Ricci et al. proposed a technique for segmentation of blood vessels using orthogonal line gradient
detectors and support vector machines. Both supervised and unsupervised learning was performed for the
detection of all blood vessels. Soares et al. used Gabor wavelet at different scales for feature extraction
along with Bayesian classifier for classification of extracted vessels. Staal, Abràmoff, Niemeijer et al.
(2004) proposed a method for segmentation of blood vessels from two-dimensional colored images. In
this, the ridges and patches were extracted which was from the vessel centerlines which in turn helped in
extraction of prominent features. The classification process was done by k-nn classifier. Sinthanayothin,
Boyce, Cook et al. (1999) proposed a method for the detection of optic disc and blood vessels using
multilayer neural network which consists of 200 input nodes and two output nodes.
64
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
PROPOSEd ALGORITHM
The proposed algorithm is carried out in number of steps given by:
1.
2.
3.
4.
5.
Image Preprocessing
Image Segmentation
Feature Extraction
Feature Selection and Optimization
Objective and Clinical Evaluation
Image Preprocessing
All the existing algorithms for the segmentation of blood vessels suffered from the problems
of non-uniform illumination since all the public databases available online need some type
of preprocessing for enhancing the images. Other problems include the presence of various
pathologies such as cotton wool spots, bright lesions, hard and soft exudates. The actual
accuracy of segmentation depends upon the accurate classification of vessels and non-vessels
which in turn depends upon the segmentation of all the blood vessels including both thick
and thin vessels.
The input image was taken from the publicly available online databases named as DRIVE,
STARE and CHASE databases. DRIVE (Digital Retinal Images for Vessel Extraction) consists
of 40 images in which 20 images are in training set and 20 images are in test set. In the database,
13 images are of healthy retina and 7 images are of pathological retina. Figure 1a and 1b show
the image of normal and abnormal retina of DRIVE dataset. The ground truth images are also
available which are segmented from three different observers manually. STARE (Structured
Analysis of Retina) consists of 20 images in which 10 images are of healthy retina and 10 images
of pathological retina. The ground truth images of two observers are also present. CHASE_DB1
(Child Heart and Health Study in England Set) consists of 28 fundus images of 14 school children
by images both eyes of each child. The reference of the results is given by two observers by giving
results of manual segmentation.
Extraction of Green Channel
The input image is RGB image, but only green color of the image has the higher contrast of blood
vessels. The red channel of the image has the problem of over saturation and there is no effective
Figure 1a. Normal retina of DRIVE dataset
65
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
Figure 1b. Abnormal retina of DRIVE dataset
information in the blue channel. So, the first step in the preprocessing step is the extraction of green
channel. Figure 2a, 2b, 2c and 2d shows the input image, corresponding red channel, green channel
and blue channel.
Conversion Into Gray Level
The green channel of the RGB image is converted into gray level as gray level images requires less
computation and less complex. Figure 3a and 3b shows the green channel and corresponding gray
scale image.
Figure 2a. Input image
Figure 2b. Red channel
66
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
Figure 2c. Green channel
Figure 2d. Blue channel
Figure 3a. Green channel
Contrast Enhancement
The appearance of the images can be increased by contrast of image by using Contrast Limited
Adaptive Histogram Equalization (CLAHE) which enhances the image locally by dividing them into
tiles. Figure 4a and 4b show the input gray level image and corresponding CLAHE image.
Image Filtering
The image needs to be filtered using the gradient directional features. This filtering process will
depict the edge strength. Figure 5a and 5b show the corresponding input image and filtered image.
67
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
Figure 3b. Gray level image
Figure 4a. Input image
Figure 4b. CLAHE image
68
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
Figure 5a. Input image
Figure 5b. Filtered image
Image Segmentation
The segmentation of retinal blood vessels is done for analyzing the flow of blood in human eye
retina. The segmentation process gives the best results if all the artifacts from them have been
removed before the segmentation process. The preprocessing step of the proposed algorithm
has extracted the green channel, enhanced the contrast and removed all the unwanted noise from
the input image. There are number of methods by which segmentation can be performed which
include the name of operators like Prewitt, Robert, Canny, different thresholding techniques,
and edge-based detection techniques.
In the proposed algorithm, thresholding-based clusters were used for the segmentation process
or for the searching of pattern of edges. This process of segmentation will make the clusters of the
same patterns and then the edges will be detected. The initial seed points were chosen by comparing
the values of intensities as the pixel with the highest intensity value will be the starting point. The
membership value of each pixel is calculated and then it is assigned to specific cluster. Figure 6 shows
the corresponding preprocessed image given as input both one normal and abnormal image and the
corresponding result of segmentation.
69
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
Figure 6a. Normal image
Figure 6b. Segmented image
Figure 6c. Abnormal image
70
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
Figure 6d. Segmented image
Feature Extraction
The quantitative information is detected in the process of feature extraction. The features of all the
segmented images need to be extracted so that the proposed system can be trained for detection of
blood vessels and for differentiating between vessels and non-vessels. Features are generally defined
as attributes or properties of various segmented pixels. The feature vector for both the normal and
pathological image pixels is made so that the individual results can be maintained. Also, the input
image has number of features; the process of feature extraction will reduce the dimensions of the
extracted features.
The process of feature extraction is done by using Independent Component Analysis (ICA) which
is a statistical technique for analysis of data by computing Independent Components (ICs) of raw
data. ICA gives the best results as it first removes the correlation between the data. The advantage
of using ICA is that it generates statistically and non-Gaussian ICs which are not variable to location
and change in phase.
Algorithm 1. Independent Component Analysis (ICA)
Input: Segmented Image
Output: Extracted features for both healthy and pathological retina.
Step 1: Calculate mean and covariance matrix of the image.
Step 2: The Independent Components of the raw feature vector will
be extracted after performing
Step 2.1: The Centering and Whitening version of the input
matrix is calculated.
Step 2.2: The features were extracted at random points
after giving them random weights and adjusting
it accordingly. The weights are adjusted by
computing the values of Negentropy which is
modified version of Entropy.
Step 2.3: The transformation matrix of the values is
computed by decorrelating the weight matrices
using decomposition method.
Step 3: The ICs computed in Step 2 are saved separately for
healthy and pathological retina. Figure 7 and 8 shows the
feature vector plot for both healthy and pathological retina.
71
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
Figure 7. Plot of feature vector for healthy retina
Figure 8. Plot of feature vector for pathological retina
Feature Optimization
The step of feature optimization helps in optimizing the solution from all the candidate solutions.
The objective function is defined for the problem and feature optimization will be performed after
maximizing and minimizing the objective function. The optimization algorithms are required for
dealing with image data of larger size. They only select the prominent features or optimize the data
according to their value and priority. Like in this problem, the features which will be helpful in
diagnosing DR will be selected and all other will not be shown in final image.
In the proposed algorithm, the features were optimized one by one using Particle Swarm
Optimization (PSO), Firefly Algorithm (FA), Lion Optimization Algorithm (LOA) and Entropy
based optimization algorithm. These all are bio-inspired methods of optimization. PSO predicts
72
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
the solution of any problem using the behavior of birds flocking and swarming. The best solution
is provided after computing values of velocity of the candidates. Both the local best and global best
solution is provided by PSO (Singh and Pandey, 2014). Firefly algorithm is based on the behavior of
fireflies which attract towards the brightness and the solutions are found by random walking of the
flies and insects. LOA is another bio-inspired optimization algorithm that is based on behavior of
lions. The special attraction of behavior of lion is their cooperation behavior for finding mates and
preys. Figure 9a, 9b and 9c shows the optimized plots of features after the optimization of feature
vector with the help of PSO, FA and LOA.
The proposed algorithm named as Entropy based optimized segmentation (EBOS) optimize the
extracted features are optimized using the values of entropy. Entropy is a measure of dispersion of
histogram. Entropy will define the uncertainty of any random variable. If the image is highly ordered,
then the values will be low. The entropy values for local as well as global neighborhood are calculated
and final optimized values are calculated. Figure 10 shows the optimized feature vector.
Feature Classification
During the process of feature classification, the optimized features are assigned to specific target class.
There will be two target classes in this case, one will be NORMAL class of images of healthy retina
Figure 9a. Optimization plot after PSO
Figure 9b. Optimization plot after FA
73
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
Figure 9c. Optimization plot after LOA
Figure 10. Optimization plot after EBOS
and another one will be ABNORMAL class of pathological retina. It is main phase of supervised
learning as it helps in classifying the extracted features into the target class and the corresponding
images as of healthy and pathological retina. The features in the proposed algorithm were classified
using Support Vector Machines (SVM) (Takkar Singh, & Pandey, 2017) which is a superior process of
classification using the principle of minimization of risks. The input features are divided into different
target classes using hyperplanes and the difference between hyperplane and corresponding target class.
Comparison of Results
The proposed optimization algorithm that is EBOS was compared with all the existing optimization
techniques that are PSO, FA and Lion using the parameters given by Accuracy, Sensitivity, Specificity,
Positive Predictive Value and False Positive Rate. The values of all the above parameters can be
calculated using True Positive (TP), False Positive (FP), True Negative (TN), False Negative (FN).
TP represents the pixels which are vessels in both manually segmentation as well as in segmentation
by proposed algorithm. FP represents the pixels which are non-vessels in ground truth images, but
vessels in proposed algorithm segmentation. TN represents the pixels which are non-vessels in both
74
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
manually segmentation as well as in segmentation by proposed algorithm. FN represents the pixels
which are vessels in ground truth images, but non-vessels in proposed algorithm segmentation.
Accuracy is defined as number of correctly classified pixels both vessels and non-vessels with
respect to total number of pixels. Sensitivity is defined as ratio of correctly classified vessels with
respect to total number of pixels. Specificity is defined as ratio of correctly classified non-vessels
with respect to total number of pixels. Positive Predictive Value (PPV) is the ability of any algorithm
for classification of vessels is really a vessel. False Positive Rate (FPR) is the ratio of pixels which
are detected as vessel pixel due to some error.
All the algorithms were implemented on DRIVE, STARE and CHASE_DB1 databases. Figure
11 shows the graph of accuracy which shows the comparison of Firefly, PSO, Lion and EBOS
algorithm for DRIVE database.
Figure 12 shows the graph of Sensitivity for all the four algorithms for STARE database. Figure
13 shows the graph of specificity for all for PPV and Figure 14 shows the graph of FPR value for
DRIVE and Figure 15 shows the graph of FPR value.
The objective evaluation of the algorithm states that the accuracy of EBOS algorithm is higher
than Firefly, PSO and Lion algorithm. Firefly is the lowest amongst all. The average accuracy achieved
by the proposed algorithm is 99.37% for DRIVE, 99.13% for STARE and 99.26% for CHASE_DB1,
respectively. The values of sensitivity, positive predictive value, false positive rate and specificity
are also better in case of all three databases.
CONCLUSION ANd FUTURE SCOPE
In this article, the segmentation of blood vessels of the human retina is performed by EBOS
algorithm in which the entropy of the features is calculated individually for the optimization of
the feature vector. EBOS algorithm consists of preprocessing phase, segmentation phase, feature
extraction and feature optimization, feature classification and objective evaluation. The objective
evaluation was performed using accuracy, sensitivity, specificity, false positive rate and positive
predictive value. The values of parameters exhibit that the proposed algorithm outperformed all
other algorithms. The proposed algorithm achieves the average accuracy 99.37% for DRIVE,
Figure 11. Accuracy graph
75
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
Figure 12. Sensitivity graph
Figure 13. Graph of positive predictive value
99.13% for STARE and 99.26% for CHASE_DB1 respectively. The average sensitivity is 99.83%
for DRIVE database, 99.53% for STARE and 99.12% for CHASE_DB1 respectively. EBOS
algorithm was able to perform so well for these images as it is based on entropy which classifies
and optimizes the images by calculating the disorder-ness in the segmented image. In future,
optimization of feature vector is done for achieving 100% accuracy of the segmentation. This
work can be further improved by segmenting all the symptoms of DR in the retina and predicting
76
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
Figure 14. False positive rate graph
Figure 15. Graph of specificity
the specific stage of DR. Different optimization techniques can be further applied on segmented
images for achieving accuracy.
ACKNOwLEdGMENT
The authors are thankful to IKG Punjab Technical University, Kapurthala, Punjab to give the
opportunity to do this research work.
77
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
REFERENCES
Anzalone, A., Bizzarri, F., Parodi, M., & Storace, M. (2008). A modular supervised algorithm for vessel
segmentation in red-free retinal images. Computers in Biology and Medicine, 38(8), 913–922.
Aslani, S., & Sarnel, H. (2016). A new supervised retinal vessel segmentation method based on robust hybrid
features. Biomedical Signal Processing and Control, 30, 1–12.
Franklin, S. W., & Rajan, S. E. (2014). Computerized screening of diabetic retinopathy employing blood vessel
segmentation in retinal images. Biocybernetics and Biomedical Engineering, 34(2), 117–124.
Franklin, S. W., & Rajan, S. E. (2014). Retinal vessel segmentation employing ANN technique by Gabor and
moment invariants-based features. Applied Soft Computing, 22, 94–100.
Fraz, M. M., Remagnino, P., Hoppe, A., Uyyanonvara, B., Rudnicka, A. R., Owen, C. G., & Barman, S. A.
(2012). Blood vessel segmentation methodologies in retinal images–a survey. Computer Methods and Programs
in Biomedicine, 108(1), 407–433.
GeethaRamani, R., & Balasubramanian, L. (2016). Retinal blood vessel segmentation employing image
processing and data mining techniques for computerized retinal image analysis. Biocybernetics and Biomedical
Engineering, 36(1), 102–118.
Hatanaka, Y., Samo, K., Tajima, M., Ogohara, K., Muramatsu, C., Okumura, S., & Fujita, H. (2016, March).
Automated blood vessel extraction using local features on retinal images. In Medical Imaging 2016: ComputerAided Diagnosis (p. 97852F). International Society for Optics and Photonics.
Henkind, P., & Wise, G. N. (1974). Retinal neovascularization, collaterals, and vascular shunts. The British
Journal of Ophthalmology, 58(4), 413–422.
Hoyer, P. O., & Hyvärinen, A. (2000). Independent component analysis applied to feature extraction from colour
and stereo images. Network (Bristol, England), 11(3), 191–210.
Joshi, S. R. (2015). Diabetes Care in India. Annals of Global Health, 81(6), 830–838.
Li, M., Zhenshen, M., Chao, L., Guang, Z., & Zhe, H. (2017). Robust retinal blood vessel segmentation based
on reinforcement local descriptions. BioMed Research International. doi:10.1155/2017/2028946
Lupascu, C. A., Tegolo, D., & Trucco, E. (2010). FABC: Retinal vessel segmentation using AdaBoost. IEEE
Transactions on Information Technology in Biomedicine, 14(5), 1267–1274.
Marín, D., Aquino, A., Gegúndez-Arias, M. E., & Bravo, J. M. (2010). A new supervised method for blood vessel
segmentation in retinal images by using gray-level and moment invariants-based features. IEEE Transactions
on Medical Imaging, 30(1), 146–158.
Marini, F., & Walczak, B. (2015). Particle swarm optimization (PSO). A tutorial. Chemometrics and Intelligent
Laboratory Systems, 149, 153–165.
Mustafa, W. A., Haniza, Y., & Wahida, K. (2017). Combination of gray-level and moment invariant for automatic
blood vessel detection on retinal image. Journal of Biometrics. Biomaterials and Biomedical Engineering, 34,
10–19.
Osareh, A., & Shadgar, B. (2009). Automatic blood vessel segmentation in color images of retina. Indian Journal
of Science and Technology, 33(B2), 191.
Rahim, S. S., Jayne, C., Palade, V., & Shuttleworth, J. (2016). Automatic detection of microaneurysms in colour
fundus images for diabetic retinopathy screening. Neural Computing & Applications, 27(5), 1149–1164.
Rajput, M. Y., Manza, R. R., Patwari, M. B., Rathod, D. D., Borde, P. L., & Yannawal, P. L. (2015). Detection
of Non-Proliferative Diabetic Retinopathy lesions using wavelet and classification using k-means clustering. In
Proceedings of the International Conference on Communication Networks (pp. 381-387). IEEE. doi:10.1109/
ICCN.2015.73
Ricci, E., & Perfetti, R. (2007). Retinal blood vessel segmentation using line operators and support vector
classification. IEEE Transactions on Medical Imaging, 26(10), 1357–1365.
78
International Journal of Healthcare Information Systems and Informatics
Volume 15 • Issue 2 • April-June 2020
Roychowdhury, S., Koozekanani, D. D., & Parhi, K. K. (2014). Blood vessel segmentation of fundus images
by major vessel extraction and subimage classification. IEEE Journal of Biomedical and Health Informatics,
19(3), 1118–1128. doi:10.1109/JBHI.2014.2335617
Saha, M., Naskar, M. K., & Chatterji, B. N. (2016). Detection of Diabetic Retinopathy Using the Wavelet
Transform and Feedforward Neural Network. In Advanced Computing and Systems for Security (pp. 101–109).
New Delhi: Springer; . doi:10.1007/978-81-322-2650-5_7
Shah, S. A. A., Tang, T. B., Faye, I., & Laude, A. (2017). Blood vessel segmentation in color fundus images
based on regional and Hessian features. Graefes Archive for Clinical and Experimental Ophthalmology, 255(8),
1525–1533.
Singh, A., & Pandey, B. (2014). Intelligent techniques and applications in liver disorders: A survey. International
Journal of Biomedical Engineering and Technology, 16(1), 27–70. doi:10.1504/IJBET.2014.065638
Sinthanayothin, C., Boyce, J. F., Cook, H. L., & Williamson, T. H. (1999). Automated localisation of the optic
disc, fovea, and retinal blood vessels from digital colour fundus images. The British Journal of Ophthalmology,
83(8), 902–910.
Soares, J. V., Leandro, J. J., Cesar, R. M., Jelinek, H. F., & Cree, M. J. (2006). Retinal vessel segmentation using
the 2-D Gabor wavelet and supervised classification. IEEE Transactions on Medical Imaging, 25(9), 1214–1222.
doi:10.1109/TMI.2006.879967
Staal, J., Abràmoff, M. D., Niemeijer, M., Viergever, M. A., & Van Ginneken, B. (2004). Ridge-based vessel
segmentation in color images of the retina. IEEE Transactions on Medical Imaging, 23(4), 501–509. doi:10.1109/
TMI.2004.825627
Takkar, S., Singh, A., & Pandey, B. (2017). Application of machine learning algorithms to a well defined
clinical problem: Liver disease. International Journal of E-Health and Medical Communications, 8(4), 38–60.
doi:10.4018/IJEHMC.2017100103
Tang, S., Lin, T., Yang, J., Fan, J., Ai, D., & Wang, Y. (2015). Retinal vessel segmentation using supervised
classification based on multi-scale vessel filtering and Gabor wavelet. Journal of Medical Imaging and Health
Informatics, 5(7), 1571–1574.
Wang, S., Yin, Y., Cao, G., Wei, B., Zheng, Y., & Yang, G. (2015). Hierarchical retinal blood vessel segmentation
based on feature and ensemble learning. Neurocomputing, 149, 708–717.
Xu, L., & Luo, S. (2010). A novel method for blood vessel detection from retinal images. Biomedical Engineering
Online, 9(1), 14. doi:10.1186/1475-925X-9-14
You, X., Peng, Q., Yuan, Y., Cheung, Y. M., & Lei, J. (2011). Segmentation of retinal blood vessels using the
radial projection and semi-supervised approach. Pattern Recognition, 44(10-11), 2314–2324.
Zimmet, P. Z., & Alberti, K. G. M. (2016). Epidemiology of diabetes—status of a pandemic and issues around
metabolic surgery. Diabetes Care, 39(6), 878–883.
Sukhpreet Kaur is pursuing a Ph.D from IKG Punjab Technical University in Digital Image Processing. She did her
B. Tech and M.tech in Computer Science and Engineering.
Kulwinder Singh Mann completed a Ph.D from IKG Punjab Technical University in Medical Informatics. He did his
B. Tech and M.tech in Computer Science and Engineering. Currently, he is a Professor in the IT department of
Guru Nanak Dev Engineering College, Ludhiana.
79