See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/338253404

Diabetic Retinopathy – Feature Extraction and Classification using Adaptive

Super Pixel Algorithm

Article in International Journal of Engineering and Advanced Technology · December 2019

DOI: 10.35940/ijeat.B2656.129219

CITATIONS READS

15 557

2 authors:

Rudra Kalyan Nayak subramanian balambigai

VIT Bhopal University, Madhya Pradesh Kongu Engineering College
74 PUBLICATIONS 462 CITATIONS 53 PUBLICATIONS 205 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Rudra Kalyan Nayak on 13 October 2020.

The user has requested enhancement of the downloaded file.

 International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-9 Issue-2, December, 2019

Diabetic Retinopathy – Feature

Extraction and Classification using
Adaptive Super Pixel Algorithm
Balambigai Subramanian, V. Saravanan, Rudra Kalyan Nayak, T. Gunasekaran, S. Hariprasath

 opthamalogists which is a time consuming process and

Abstract: Diabetic Retinopathy is an ocular manifestation of sometimes delay in diagnosis may lead to delay in treatment
diabetes . The longer a person has diabetes, higher are the to patients.
chances of having diabetic retinopathy in their visual system. Diabetes is usually caused when insulin is produced less in
Hence the objective of this research work is to propose an body or if the human body develops resistance to glucose or
automated, suitable and sophisticated approach using image
when both the causes occur simulataneously[2].A person
processing so that diabetic retinopathy can be detected at early
levels easily and damage to retina can be minimized. A vital point with diabetes has a condition in which the quantity of glucose
of diabetic retinopathy that it causes detectable changes in the in the blood is too elevated (hyperglycemia). This is because
blood vessels of the retina. The focal blurred edges are detected so the body does not produce enough insulin, produces no
as to dismiss the false alarms. A two-level approach is used here to insulin, or has cells that do not respond properly to the insulin
classify data. Firstly, optimal features are extracted from the the pancreas produces[3]. Research says for about 90% of
training data and secondly, the classification is done by the use of diabetics, eye disease can be managed through better medical
the adaptive super pixel algorithm and then the test data is treatment and periodic checkup of the eyes. Image
analyzed. Adaptive super pixel algorithm can adjust the weights of processing is used by the ophthalmologists to diagnose the
various features based on their discriminating ability. After the
various eye diseases . In India, it is predicted that 79.4 million
application of algorithm, the diabetic eye is detected by means of
various parameters like colour, texture, spatial distance, contour, people may suffer from diabetes, out of which about 75
mean, standard deviation, entropy and maximum pixel points. percent of people those who have this disease for more than
This research can aid the doctor for easy detection of the disease 20 years may have certain form of problem in vision due to
as it given an accuracy of about 98.33%. diabetic retinopathy[4] . The blood vessel changes in the
Keywords: Retinopathy, fundus, adaptive super pixel, retina of the eye gives information about change in the values
classification of some of the features of the eye, such as blood vessels of
the eyes , optic disc (OD), that are helpful to find the
I. INTRODUCTION occurrence of diabetic retinopathy (DR) or any other eye
diseases.
D iabetes occurs among various people due to various Poor people are easily affected by diabetic retinopathy due
to the non availability of screening , early diagnosis and
factors like sedentary life style , stress and eating habits.
Diabetic Retinopathy usually happens if a person has treatment. Many screening tools are available to diagnose
diabetes for more than ten years or more. Diabetic diabetic retinopathy. Fundus cameras capture digital image
Retinopathy (DR), causes around five percent of blindness of the blood vessels in the retina and sometimes they may be
among people and this is one of the major reason that causes having excessibe brightness etc that reduce the quality of the
blindness among diabetics. According to World health image. Hence, there is a need to enhance the images by the
Organization (WHO) estimation, 425 million of world use of image enhancement. Features of the eye that help to
population is having the diabetes [1] .Research indicates that detect the retinopathy are blood vessels, textures
progress of the reduction of can be reduced or eradicated if microaneurysms , exudes etc. These features increase in size
diabetic retinopathy is found at an early stage. It is found that and may destruct the capillaries causing blood leakage in the
large number of people having diabetic retinopathy are being eye leading to blindness.. Hence, the objective of this
tested for this disease by manual method by trained research is to detect diabetic retinopathy earlier to minimize
the damage to retina[5].

Revised Manuscript Received on December 15, 2019. II. LITERATURE REVIEW

* Correspondence Author
Dr. Balambigai Subramanian, Associate Professor, Department of Ronakorn Panjaphongse (2017) used the morphological
Electronics and Communication Engineering, Kongu Engineering College. operators along with support vector machine (SVM) to
E-mail: [email protected] screen diabetic retinopathy earlier among the diabetics.
Dr. V. Saravanan, Associate Professor, Department of Computer
Science and Engineering, K L Deemed to be University, Guntur (Andhra Initially, pre processing of the image is done to improve
Pradesh) India. image quality,followed by filtering after edge detection and
Dr. Rudra Kalyan Nayak, Associate Professor, Department of morphological operations. The extracted features are given as
Computer Science and Engineering, K L Deemed to be University, input to the support vector machine classifier which
Vaddeswaram (Andhra Pradesh) India.
Dr. T. Gunasekaran, Department of Electronics and Communication differentiates between healthy
Engineering, Higher College of Technology, Muscat
Mr. S. Hariprasath ,Assistant Professor in Department of Electronics
and Communication Engineering, Saranathan College of Engineering,
Trichy (Tamil Nadu) India.

Published By:
Retrieval Number: B2656129219/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.B2656.129219 618 & Sciences Publication
 Diabetic Retinopathy – Feature Extraction and Classification using Adaptive Super Pixel Algorithm

and diabetic retinopathy affected eye images. The proposed 3.2 SUPPORT VECTOR MACHINE
method iwas verified on three datasets and this method gave Support Vector Machine used points in space to represent
an accuracy of 90% . The grading of severity was based on the features so that the data can be mapped into different
the values of the various above mentioned parameters for the categories with clear boundary gap between them [14].
detected lesions [6]. SVM is more suitable to analyze the fundus retinal image and
Manoj Kakarla (2016) discussed about extracting the to classify them into various classes such as normal or
various features of the eye in an optimal manner, followed by non-proliferative diabetic retinopathy or proliferative
feature extraction and classification of diabetes using naïve diabetic retinopathy. Using SVM classifier that uses a non
bayes classifier. This work discusses an expert system to linear kernel function, the accuracy level increases due to the
identify the stage of diabetes based on patients test result. techniques used for extraction and classification.
Before performing classification process, optimal features
are extracted based on data sets for the purpose of reducing 3.3 K-NN CLASSIFIER
time complexity over the classification process[7]. Then the K-Nearest Neighbour (KNN) algorithm classifies an object
classification process is performed by naïve bayes theorem to an appropriate class based on the K value. Generally, K
to identify each patient’s stage in the progress of the disease value is considered to be one for classification of retinal
diabetes. By implementing these concepts, the efficiency of image. The graphs indicate the “variation” between the
classification increases and also reduces the time complexity statistical parameters, which is the base for a K-NN classifier
for performing the classification process. system to find out nearest neighbour for a new image [20].
K. R. Ananthapadmanaban etal (2014) proposed the This variation is found better for the statistical parameters
chance of occurrence of diabetic retinopathy based on extracted by using K-means classifier and this is best suited
classification using data mining methods[8]. In this work, for classifying a retinal image as healthy or abnormal.
diabetic retinopathy has been diagnosed earlier. The
preprocessing operation improves quality of the image by IV. PROPOSED SYSTEM
reducing the effects of excessive brightness. In the second
step, the optic disc has to be located and extracted from the 4.1 PROPOSED METHODOLOGY
In this proposed work, the input images of size 256x256
retinal image. The third step consists of the segmentation to
pixels are pre-processed to convert the input colour images in
find the exudates . In the last step, the neural network is used
RGB to gray images because inherent complexity of gray
for feature extraction. The final results were compared with
image is less than the colour images. Also, gray scale reduces
the opinion gathered from an eye doctor and it was found that
the specificity was 95% and sensitivity was 96.65%. the complexity to one dimension from three dimension. RGB
Hilary W. Thompson (2016) proposed a novel method segmentation is used for dividing image into several region
with their own characteristics for extracting useful target like
based on hierarchical decomposition and post filtering of
lines and curves of an image. Image enhancement allows to
edges to detect blood vessels to detect the occurrence of
brighten an image and makes it easier to identify the key
diabetic retinopathy at an early stage. This algorithm reduced
features in preprocessing. The block diagram which is used
the false decisions and was quite faster than conventional
methods and needed less storage requirements[9].Ms. for this research is shown in figure 4.1.
Bhaminee R Shetty (2017) proposed the automatic
segmentation of microaneurysm using attribute filtering to
reduce noise and thereby detected the exudates.Later, the
feature of exudates was extracted from green channel of the
RGB retinal image which was later given as input to the
classifier. Statistical parameters such as mean, standard
deviation, kurtosis etc were considered for classification and
SVM gave sensitivity of 96.9%.[10]

III. EXISTING SYSTEM

3.1 NAÏVE BAYESIAN CLASSIFIER
This method extracts the optimal features initially from the
training data [19]. Later, the values of positive and negative
probability is found .Later, a new dataset of same size is
created to which the values of previous data set are
forwarded to help in the classification process. After
completion of optimal feature extraction process, that dataset
is taken and sent to classification. By applying classification
algorithm, each patient’s stage over the diabetes can be
identified. The probability formula for naïve bayes algorithm
is shown in equation 3.1
Figure 4.1 Block diagram of proposed system

……... (3.1)

Published By:
Retrieval Number: B2656129219/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.B2656.129219 619 & Sciences Publication
 International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-9 Issue-2, December, 2019
Segmentation of preprocessed images is done by using 4.2 ADAPTIVE SUPER PIXEL ALGORITHM
DWT (Discrete Wavelet Transform) using Haar family Super pixel segmentation uses super pixels that show clear
wavelets. The preprocessed image is classified into two types entity for human eye.This algorithm leads to cost reduction
based on their nature by Hybrid Structure Feature (HSF) and as super pixels are used and processing of super pixels
Hybrid Texture Feature (HTF). The HSF performs optical become easier. Boundary are clearly marked based on
disc removal, threshold point calculation. And to find the regularity of the super pixels .Design of adaptive super pixel
region of interest. HTF involves in blood vessel removal and algorithm is due to the following factors such as colour
dots part detection.In feature extraction the black points, difference is not sufficient to produce meaningful entity in
white points, mean error, standard deviation, entropy and low contrast images and various colours have various values
maximum pixel points are extracted. for its parameters of texture, colour, contour etc.
The flow chart of the above algorithm is shown in figure 4.2.

Figure 4.2 Flow chart of adaptive super pixel algorithm

The novelties of adaptive super pixel lie in: 4.5 IMAGE SEGMENTATION
• Consistency of Perception Image segmentation divides the image into many segments
• Adaptability of content to simplify or change its representation so that its analysis
• Simplicity and efficiency becomes easier. Labels are assigned so that same label
contains the pixels with similar characteristics. The steps
4.3 THE FEATURES AND DISTANCE MEASURE involved are
Colour difference dc: It is calculated using equation 4.1 as i) Converting the RGB image to grayscale to retain
referred in [23] luminance and remove hue and saturation.
………… (4.1) ii) Improving the contrast by the method of histogram
equalization.
where l, c, and h represents the differences of l, c, iii) Next, image enhancement is done by contrast adjustment,
and h between two pixels morphological filtering so as to Return a modified version
• Significance: Colour difference is used to find the of the original retinal image.
contrast of each pixels. iv) Optic disk present on the retina is very bright and it has to
Spatial distance: be extracted using feature extraction methods to identify
• Significance: Spatial distance is used to find the the blood vessels.
length of a straight line in a 3-D space i.e. the distance
v) Blood vessels are extracted to be examined for identifying
between the two-pixel points is calculated.
various eye diseases such as diabetic retinopathy.
Image gradient:
• Significance: Image gradient is useful to find the vi) The next step is the classification of exudates into hard or
changes in directional intensity of an retinal image . soft exudates which is done on the basis of threshold
Weber local descriptor: value.
• Significance: Weber local descriptor finds patterns and vii) Mean, Median, Standard deviation are calculated and
difference in texture to find the features. based on these features, the adaptive super pixel algorithm
Distance D between pixels is shown in equation 4.2. classifies the image of the eye as normal or diabetic eye.

….(4.2)

Published By:
Retrieval Number: B2656129219/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.B2656.129219 620 & Sciences Publication
 Diabetic Retinopathy – Feature Extraction and Classification using Adaptive Super Pixel Algorithm

V. RESULTS AND DISCUSSION RGB segmentation is a process in which an image is

divided into different regions in order to isolate the areas of
5.1. RESULTS AND DISCUSSION
interest on it.The RGB segmentation of normal eye is shown
Data Set Information: in figure 5.3.
Dataset used is Messidor image for this research work. All
features represent either a detected lesion, a descriptive
feature of an anatomical part or an image-level descriptor.
The data set comprises 60 images including 25 diabetic
retinopathy images and 35 healthy retinal images of size 256
x 256 pixels.
The input image 11 of a normal eye is taken from above data
set and is shown in figure 6.1 is taken for processing.

Figure 5.3 RGB segmentation of normal eye

5.1.5 Optic Disk Location

The optic disk location of normal eye is shown in figure 5.4

Figure 5.1 Normal eye

5.1.2 Enhanced Image
Image enhancement is accomplished by the use of filters,
image editors and other tools to change the properties of
retinal image of an eye.The enhanced image of normal eye is
shown in figure 5.2.

Figure 5.4 Optic disk location of normal eye

The white point represents the location of optic disk in the
image.

5.1.6 Region of Interest

In region of interest the boundaries of defected lesions of
eye may be defined on an image or in a volume, for the
purpose of measuring its size. The region of interest for red,
Figure 5.2 Enhanced image normal eye green and blue channel of normal eye is shown in figure 5.5
5.1.4 RGB Segmented Image

Figure 5.5 Region of interest of normal eye

The green channel is used for its good contrast when

compared with red and blue channel.

5.1.7 Optic Disk Removal

The green channel is taken for further processing from
region of interest .The optic disk removed image of normal eye
is shown in figure 5.6

Published By:
Retrieval Number: B2656129219/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.B2656.129219 621 & Sciences Publication
 International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-9 Issue-2, December, 2019
After optic disc removal the blood vessel trace is identified
and removed for further classification
5.1.9 Dots Part
The white points and black points of normal eye is shown in
figure 5.8. The occurrence of microaneurysm in eye is
displayed as dots part.

Figure 5.6 Optic disk removed image of normal eye

The optic disk is removed for better enhancement and

classification
5.1.8 Blood Vessel Extraction
Blood vessel extraction is a processing step to extract
vessels away from image to investigate the existence on
some disease. The extracted blood vessel of normal eye is Figure 5.8 Dots part of normal eye
shown in figure 5.7.
Based on the number of white points the eye is classified as
normal and affected.
5.1.10 Parameters measured
The parameters like mean, standard deviation, entropy
and maximum pixel points of normal eye are shown in
table 5.1.

Figure 5.7 Blood vessel extracted image of normal eye

Table 5.1 Parameters of normal eye (Image 1)

Image Maximum Standard Affected Normal pixel
number Deviation pixel point Entropy deviation pixel point point
11 176.358 197.858 1.405 60.738 2.790 91422.30

5.2 RESULTS OF AFFECTED EYE 5.2.2 Enhanced Image

The enhanced image of affected eye is shown in figure 5.10
5.2.1 Input Image
An affected eye is shown in figure 5.9 is taken for
processing.

Figure 5.10 Enhanced image of affected eye

Figure 5.9 Image of affected eye

Published By:
Retrieval Number: B2656129219/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.B2656.129219 622 & Sciences Publication
 Diabetic Retinopathy – Feature Extraction and Classification using Adaptive Super Pixel Algorithm

5.2.4 RGB Segmentation

The RGB segmentation of affected eye is shown in figure
5.11

Figure 5.12 Optic disk location of affected eye

Figure 5.11 RGB segmentation of affected eye The white point represents the location of optic disk in the
The enhanced image is differentiated by RGB image.
segmentation
5.2.5 Optic Disk Location 5.2.6 Region Of Interest
The optic disk location of affected eye is shown in figure The region of interest for red, green and blue channel of
5.12. affected eye is shown in figure 5.13.

Figure 5.13 Region of interest of affected eye

The green channel is used for its good contrast when

compared with red and blue channel.
5.2.7 Optic Disk Removal
The Optic disk removed image of affected eye is shown in
figure 5.14.

Figure 5.15 Blood vessel extracted image of affected eye

After optic disc removal the blood vessel trace is identified

and removed for further classification

5.2.9 Dots Part

Figure 5.14 Optic disk removed image of affected eye The white points and black points of affected eye is shown
in figure 5.16
The green channel is taken for further processing from
region of interest. The optic disk is removed for better
enhancement and classification
5.2.8 Blood Vessel Extraction
The extracted blood vessel of affected eye is shown in
figure 5.15.

Published By:
Retrieval Number: B2656129219/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.B2656.129219 623 & Sciences Publication
 International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-9 Issue-2, December, 2019
5.2.10 Parameters measured
The parameters of affected eye are shown in table 5.2.
Table 5.2 Parameters of affected eye (Image 045)
Affected Normal
Image Maximum Standard pixel pixel
number Deviation pixel point Entropy deviation point point
Image
045 165.289 191.877 1.414 65.705 2590.582 88834.516

5.3 FINAL CLASSIFICATION AND RESULT

By analyzing all the features and parameters the results are
Figure 5.16 Dots part of affected eye displayed as “INPUT IMAGE NORMAL (or) “INPUT
IMAGE AFFECTED”.The final result of normal is shown
Based on the number of white points the eye is classified as in figure 5.17.
normal and affected.

Figure 5.17 Final result of normal eye

The final result of affected eye is shown in figure 5.18

Figure 5.18 Final result of affected eye

In normal eye blood vessel there is no growth of blood 5.4 PARAMETERS

vessel and no blood leakage. 5.4.1 Normal eye
Where as in abnormal eye there is more amount of blood Various parameters of all normal eye image in data set is
leakage and emergence of new blood vessels. shown in table 5.3.

Published By:
Retrieval Number: B2656129219/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.B2656.129219 624 & Sciences Publication
 Diabetic Retinopathy – Feature Extraction and Classification using Adaptive Super Pixel Algorithm

Table 5.3 Parameters of all normal eye image in data set

Ranges of parameters of normal eye : 5.4.2 AFFECTED EYE

Standard deviation lies in the range between 55 - 63. Various parameters of all affected eye images in data set is
Affected pixel point lies in the range between 0 – 290 shown in table 5.4.
Normal pixel point lies in the range between
91000 – 92000 as per reference paper[9].

Table 5.4 Parameters of all affected eye images in data set

Ranges of parameters of affected eye : 7 62.555 4.185 91420.912 YES

Standard deviation lies in the range between 63-75. 10 62.575 2.79 91422.307 YES
Affected pixel point lies in the range of 290 and above
Normal pixel point lies in the range of 91000 and below as Image 004 62.391 0 91425.098 YES
per reference paper[9]. Image 047 52.093 115.788 91309.31 YES
Image 040 60.644 29.295 91395.802 YES
5.5 DOCTOR’S OPINION FOR NORMAL EYE 8 61.807 16.74 91408.3 YES
Doctor’s confirmation for normal eye is shown in table 5.5
11 60.738 2.79 91422.321 YES
Table 5.5 Doctor’s verification for normal eye 12 62.551 4.185 91420.912 YES
AFFECTED NORMAL DOCTOR' 14 62.559 754.714 90670.383 NO
IMAGE STANDARD PIXEL PIXEL S 15 62.082 2.79 91422.307 YES
NUMBER DEVIATION POINT POINTS OPINION
1 60.738 2.791 91422.311 YES
4 62.647 4.185 91420.913 YES

Published By:
Retrieval Number: B2656129219/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.B2656.129219 625 & Sciences Publication
 International Journal of Engineering and Advanced Technology (IJEAT)
ISSN: 2249 – 8958, Volume-9 Issue-2, December, 2019

16 61.484 122.763 91302.334 YES REFERENCES

18 59.7111 8.3702 91416.727 YES 1. https://www.allaboutvision.com/en-in/conditions/diabetic/
2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2613584/
21 62.976 57.1964 91367.901 YES 3. https://www.niddk.nih.gov/health-information/diabetes/overview/wha
23 51.44 216.02 91206.077 YES t-is-diabetes
4. https://www.iapb.org/wp-content/uploads/Guidelines-for-the-Compre
24 62.929 21.9 91397.197 YES
hensive-Management-of-DR-in-India.pdf
Image 003 62.201 219.02 91206.077 YES 5. Salz, D. A., & Witkin, A. J. (2015). Imaging in diabetic
Image 043 61.398 0 91425.098 YES retinopathy. Middle East African journal of ophthalmology, 22(2),
145–150. doi:10.4103/0974-9233.151887
6. Sreng, Syna, Noppadol Maneerat, Kazuhiko Hamamoto and Ronakorn
5.6 DOCTOR’S OPINION FOR AFFECTED EYE Panjaphongse. “Automated Diabetic Retinopathy Screening System
Doctor’s confirmation for affected eye is shown in table 5.6 Using Hybrid Simulated Annealing and Ensemble Bagging Classifier.”
(2018). DOI:10.3390/app8071198
7. Manoj Kakarla (2016), “Optimal Feature Extraction and Classification
Table 5.6 Doctor’s verification for affected eye of Diabetes using Naïve Bayesian Classifier”, International Journal of
AFFECTED NORMAL DOCTOR'S Engineering Trends and Technology (IJETT), Vol-42, pp 63-70.
IMAGE STANDARD
NUMBER DEVIATION
PIXEL PIXEL CONFIRMA 8. K. R. Ananthapadmanaban and G.Parthiban (2014), “Prediction of
POINT POINTS TION YES Chances - Diabetic Retinopathy using Data Mining Classification
9 69.106 756.109 90668.988 YES Technique”, Indian Journal of Science and Technology, Vol 7, pp
1498–1503.
13 68.995 1037.906 90387.191 YES 9. S, Kandiraju N, Thompson HW. Design and implementation of a
68.881 3648.019 87777.078 YES unique blood-vessel detection algorithm towards early diagnosis of
diabetic retinopathy. In: International conference on information
Image 013 technology: coding and computing, vol. 1. 2005. p. 26–31.
67.246 5868.917 85556.18 YES 10. http://www.kscst.iisc.ernet.in/spp/39_series/SPP39S/02_Exhibition_P
rojects/177_39S_BE_1679.pdf
Image 042 11. A. B. Chavan (2017), “ Image Processing Analysis on Retina Blood
68.141 754.714 90670.383 YES Vessel for Detecting Glaucoma”, International Journal of Electronics,
Communication & Soft Computing Science and Engineering, Vol. 45,
Image 005 Issue 1, pp 63-85.
20 63.181 1343.419 90754.085 YES 12. Abramoff, M.D and Garvin, M.K. (2010), “Retinal Imaging and Image
Analysis”, IEEE Reviews in Biomedical Engineering, Vol. 3, pp
66.99 636.136 90788.961 YES 12-20.
Image 044 13. Aiswarya Iyer (2015), “Diagnosis of diabetes using classification
mining techniques”,International Journal of Data Mining &
3 70.851 412.93 91012.167 YES Knowledge Management Process Vol.5, pp 52-57.
12 63.493 341.783 91083.314 YES 14. C. E. Marwan D. Saleh (2012), "An automated decision support system
for non-proliferative diabetic retinopathy disease based on MAs and
17 63.872 313.883 91111.21 YES HAs detection", Computer Methods and Programs in Biomedicine,
20 63.181 1343.419 90754.085 YES Vol. 108, pp. 186-196.
15. G. D. Joshi, J. Sivaswamy, and S. R. Krishnadas (2011), “Optic Disk
22 63.913 671.012 90754.085 YES and Cup Segmentation from Monocular Colour Retinal Images for
Image 035 68.354 661.247 90763.85 YES Glaucoma Assessment,” IEEE Trans. Medical Imaging, Vol. 30, no. 6,
pp. 1192–1205.
Image 015 64.192 669.617 90755.48 YES 16. Huiling Chen (2017), “Automatic Analysis of Microaneurysms
Image 006 68.796 749.134 90675.963 YES Turnover to Diagnose the Progression of Diabetic Retinopathy”,
Institute of Electrical and Electronics Engineers, Vol - 8, pp 976-997.
Image 007 67.031 694.728 90730.369 YES 17. Kevin Noronha (2015), “Support System for the Automated Detection
Image 016 69.756 725.418 90699.678 YES of Hypertensive Retinopathy using Fundus Images”, International
Conference on Electronic Design and Signal Processing, Vol. 19, pp
Image 042 67.246 5868.917 85556.8 YES 567-589.
59 images out of 60 images gives the exact result. Thus, the 18. Mohammed Ghazal (2016), “Early Diagnosis of Diabetic Retinopathy
in OCTA Images Based on Local Analysis of Retinal Blood Vessels
accuracy of the project is 98.33% and Foveal Avascular Zone”, International Conference on Pattern
Recognition (ICPR), Vol.5, pp 623-647.
VI. CONCLUSION AND FUTURE WORK 19. Muhammad Nauman Zahoora (2018), “Fast Optic Disc Segmentation
in Retina Using Polar Transform”, International Journal of Engineering
Screening of diabetic retinopathy automatically is very Trends and Technology (IJETT), Vol – 6, pp 12-18.
difficult and it is important as it helps the eye specialists to 20. Syna Sreng and Noppadol Maneerat (2018), “Primary Screening of
Diabetic Retinopathy Based on Integrating Morphological Operation
diagnose the various grades of diabetic retinopathy during the
and Support Vector Machine”, International Conference on Intelligent
examination of the eye which is accomplished by the Informatics and Biomedical sciences, Vol – 3, pp 1– 3.
combination of various steps such as the integrated use of 21. T. Teng, et al. (2002), "Progress towards automated diabetic ocular
morphological operations along with adaptive super pixel screening: a review of image analysis and intelligent systems for
diabetic retinopathy", Medical and Biological Engineering and
algorithm. This helps doctors to cure patients at the earlier
Computing, Vol. 40, pp. 2-13.
stage and also prevents eye transplantation. The severity of 22. Tamilselvan, KumaravelSubramaniam, and GovindasamyMurugesan.
the diabetic retinopathy can be further graded based on the "Survey and analysis of various image fusion techniques for clinical
detected lesions and their quantities. The performances of the CT and MRI images." International Journal of Imaging Systems and
Technology vol.24, no.2, pp.193-202, 2014.
method were also measured by specificity 95% and accuracy
23. Xiaolin Xiao, Yue-Jiao Gong, Yicong Zhou. "Adaptive superpixel
98.33%.In the future work, identifying more features that will segmentation aggregating local contour and texture features", 2017
improve the performance of adaptive super pixel algorithm.. IEEE International Conference on Acoustics, Speech and Signal
The accuracy can also be increased by improving the Processing (ICASSP), 2017
performance of the data or even by algorithm tuning. In
addition, this system can be used to detect other retinal
diseases.

Published By:
Retrieval Number: B2656129219/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.B2656.129219 626 & Sciences Publication
 Diabetic Retinopathy – Feature Extraction and Classification using Adaptive Super Pixel Algorithm

AUTHORS PROFILE
Dr. S. Balambigai received her B.E. degree in ECE from
Bharathiar University with distinction, M.E. (Applied
Electronics) and Ph.D from Anna University, Chennai.
She has around 35 publications in reputed national and
international and National journals/conferences and three
patent to her credit. She has a total of 16 years experience
and is currently working as an Associate Professor in the Department of
Electronics and Communication Engineering in Kongu Engineering College.
Developments in biomedical engineering and computer networks keeps her
fascinated to carry on further research in the these domains.

Dr. V. Saravanan is currently working as an Associate

Professor in the Department of Computer Science and
Engineering, K L Deemed to be University,Koneru
Lakshmaiah Education Foundation, Guntur,Andhra
Pradesh, India. He completed M.E. and Ph.D.degrees in
Computer Science and Engineering from
AnnaUniversity, Chennai, Tamilnadu, India. He has published 20+ Journal
papers and 5+ patents. His research interest includes Wireless Networks,
Mobile Computing, and Security.

Dr. Rudra Kalyan Nayak is presently working as

Associate Professor in the Department of Computer
Science & Engineering at KoneruLakshmaiah
Education Foundation (Deemed to be University),
Green Fields, Vaddeswaram, Andhra Pradesh, India.
He has got his M.Tech in Information Technology and
Ph.D in Computer Science & Engineering from Siksha ‘O’ Anusandhan
(Deemed to be) University, Odisha, India. He has together more than 10
years of teaching, mentoring and research experience. His research interest
lies in the field of Artificial Intelligence, Financial Engineering,
Bioinformatics and Computer Vision. He is professional member of ISTE
and ACM.

Dr. T. Gunasekaran received his B.E. degree in ECE

from Bharathiar University , M.E. (Communication
Engineering) from Birla Institute of Technology and
Science, Pilani and Ph.D (Information and
Communication Engineering) from Anna University,
Chennai. He has around 22 publications in reputed
National and International journals/conferences and one patent to his credit.
He has a total of 19 years experience in teaching & research and is currently
working as Program Director in the Department of Engineering- EEE
Section, Electronics and Communication Engineering in Higher College of
Technology, Muscat, Sultanate of Oman. His area of interest includes
Antenna design

S. Hariprasath has received B.E degree in 2000 and

Masters in Communication Systems in 2008. He has
published papers in 8 IEEE conferences, 7 International
conferences. He has published 1 paper in national
journal, 4 papers in international journals (annexure 2),
and 1 international journal publication. He has
published 1 paper in Science Citation Indexed Journal
(SCI) to his credit. He is currently pursuing research in
Pattern Recognition in VLSI. He has 16 years of teaching experience. He has
guided more than 14 PG projects and 30 UG projects in the areas of Image
processing, VLSI and Robotics. He has received Special Prize for
contribution in eYRC15. His areas of interest include Biometrics, Pattern
Recognition, FPGA Programming. He works as Assistant Professor in
Department of Electronics and Communication Engineering , Saranathan
College of Engineering ,Trichy.

Published By:
Retrieval Number: B2656129219/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijeat.B2656.129219 627 & Sciences Publication
View publication stats