Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Diabetic Retinopathy Diagnosis using Second Order

Edge Detection
1 2
Satish Kumar Kushwaha Dr. Neelesh Jain
M.Tech Research Scholar Professor
Dept. of Computer Science &Engineering Dept. of Computer Science & Engineering
SAM College of Engineering & Technology, Bhopal, SAM College of Engineering & Technology, Bhopal,
Madhya Pradesh, India Madhya Pradesh, India

3
Shekhar Nigam
Asst. Professor
Dept. of Computer Science & Engineering
SAM College of Engineering & Technology, Bhopal, Madhya Pradesh, India

Abstract:- When a person has diabetic retinopathy (DR), I. INTRODUCTION

even after having the condition for a long time, they are
highly unlikely to be aware of it. Not everyone is really The delicate retinal tissues are directly impacted by
familiar with this illness. This illness is a little different diabetic retinopathy, which causes the blood vessels to
from others since, depending on the diagnostic enlarge and begin to leak fluid. A person may get partial or
syndrome, every diabetes patient has a risk of total vision loss as a result of diabetic retinopathy. A person
developing diabetic retinopathy. Various studies have who has had diabetes for a long period may have the highest
been done in this area, but a good method is still risk of developing diabetic retinopathy. This illness can only
needed. A neural network in machine learning needs to be healed and made to quit damaging the cells; it cannot be
be trained very well because if it isn't, the system won't treated [1]. For diabetic individuals, routine testing is crucial
be able to provide decent results. The rate of false for detecting the disease at an early stage. To categorise the
alarms is higher due to poor training. However, there is blood vessels or backdrop of the exudates, a method must be
another method—an edge detection tool—by which DR effective. Because blood vessels can also manifest as deficits
may be detected more accurately. Edge has the ability to and exudates, it might be challenging to recognise the right
extract the geometry of impairments, and the density of diagnosis if the system is not skilled. High-precision DR
the retrieved region determines whether or not it is diagnosis is possible by image processing. The number of
diabetic retinopathy. The exudates from the fundus diabetic people suffering from diabetic retinopathy has
picture are extracted by the proposed method using the increased. One of the most common illnesses, DR is the
Sobel Edge Detection tool. Prior to that approach, a primary cause of vision loss in people in their middle years.
colour mapping tool was used to make exudates from the
fundus picture more visible. A colour mapping tool helps Little alterations in the retinal vessels are how DR
improve the visibility of some patches that the illness develops. Microaneurysms are the main differentiable
may cause. The backdrop can also be classified by aberrations. The intraregional drain is produced as a result of
changing the colours such that exudates are more the twisted microaneurysms. This starts the first stage of
obvious than in the original image. The suggested system diabetic retinopathy (DR), also known as mild non-
has more accuracy than the existing model and is tested proliferative diabetic retinopathy. The microinfarcts in the
using the Messidor benchmark. retina obstruct the veins over time as the retinopathy
worsens. Delicate exudates are the medical term for these
Keywords:- Diabetic Retinopathy, Fundus Image, Sobel little infarcts. This kind of retinopathy is known as severe
Edge Detection, Color Mapping, Retina, Optic Cup. non-proliferative diabetic retinopathy when more than three
abnormalities are present at the same time.

IJISRT23JUL471 www.ijisrt.com 2280

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 1 Basic Architecture of Diabetic Retinopathy Image [2]

Figure 1 depicts the retinal architecture in diabetic to combine different types of data from several photographs
retinopathy, which includes both soft and hard exudates. An into a single image [3].
picture of the fundus can be used to identify diabetic
retinopathy. The retinal region of the eye called the fundus
is visible in black. An picture of the fundus can be used to
diagnose any eye diseases. In the domain of automatically
identifying diabetic retinopathy, it is essential. Various
fundus samples from healthy and DR patient subjects are
used to train machine learning models. The machine
requires thousands of images in order to train a model that is
effective with input photographs. It increases system
expenses and processing time, nevertheless. Because it
doesn't require any form of machine learning methodology
or model, an edge detection technique might be highly
useful to substitute the machine learning-based model. Pre-
processing helps to make photos or exudates more visible
and can process input images directly. Next, edge detection
can extract the shape, size, or structural appearance of the
Fig 2 Fundus Image of the Eye with Diabetic Retinopathy
DR exudates through which a correct decision can be made
Exudates [4]
that could save lives. A novel strategy for improving the
funuds picture for subsequent proceedings is colour
The fundus picture of the eye with diabetic retinopathy
correction.
exudates is shown in Fig. 2. The diabetic retinopathy
abnormalities that are highlighted in yellow indicate the
The optical makeup of the veins or blood vessels over
degree of infection that has developed.
the retinal picture might reveal the limitations brought on by
disorders of the eye. The most frequent alterations, such as
vascular, optic circle, and fova changes, are used to examine
particular eye disorders, such as diabetic retinopathy (DR)
and other eye illnesses. Physical DR identification has been
done using a variety of screening tools. The retinal vessels
are photographed using computerized fundus screening
devices; as a result, the fundus image securing cycle can
significantly degrade the image quality. Therefore, image
improvement is always crucial to advancing the optimum
image quality. A few methods are suggested by experts to
improve the nature of retinal pictures. Image enhancement,
discontinuity, highlight extraction, and grouping are some
image management techniques that analysts use to examine
eye diseases. To spot changes in clinical pictures, image
recording is used. For successful recruitment, numerous
photos captured from diverse angles are arranged in a single
direction framework. Image combining is a technique used
Fig 3 Block Diagram of Proposed System

IJISRT23JUL471 www.ijisrt.com 2281

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
The block design of the suggested system is shown in
Fig. 3, where a dataset must first be downloaded in order to
get testing data for inference. The picture can then be
improved using colour mapping, the edges can be retrieved
using sobel edge detection, and the broken areas or edges of
the image that were missed during edge detection may be
filled with morphological dilation. Exudates can then be
divided into groups according to their density. Finally, a
choice can be made on whether or not the picture is related
to diabetic retinopathy.

II. RELATED WORKS

A segment based on shading depiction and its fusion
with the optimal colour space and Fuse C-Medium (FCM)
Numerous studies have been conducted in the area of
diabetic retinopathy, using a variety of approaches to get clustering was proposed by Alireza et al. [7]. They used
exact data on the fundus picture. Another method to locate retina shading data to achieve their goals and demonstrated
the optic circle was suggested by Ravishankar et al. [5], who advances using changes based on a dim scale. An
explicitness of 85.4, an affectability worth of 97.2, and an
had previously identified the important blood veins and used
accuracy of 85.6% were provided by the FCM grouping. In
their division to determine the approximate location of the
optic plate. Several classifiers were tested by Obscured C- this article the center of the pixel can be changed with small
Media Clustering, including SVM, neural network, PCA, window w and stated as pn. Utilising a back propagation
and generic Bayesian grouping. refers to the newly mainstream, GG Gardener et al. The regions of exudates,
vascularity, edoema, and microanurism were selected to
developed method of identifying the optic plate, which is
home to the important blood arteries and the irregular define the illness phases. It may be completed by examining
images from 100 and 47 DR patients as well as thirty
location of the optic circle caused by their division.
regular retinal images with exudates, draining, or
Additionally, it is restricted by using shade details. It
microanuria, as well as retinal images with and without
demonstrates how several components, including blood
arteries, exudates, tiny aneurysms, and drainage, may be blood vessels. The succeeding explicitness and affectability
values are, respectively, 88.4 and 83.5. Using the image
exactly tied to various changes when they are used
preprocessing and mass localization technique on retinal
appropriately. The victory rate for optical circle limitation is
images, Meher Madhu Dharmana et al. [9] established an
97.1%, the affectability is 95.7%, 95.2%, and 90.5%
microforms individually. These provide a true efficient and improved extraction procedure for obtaining
representation of these frameworks and contrast favourably fundus characteristics. On a size of 0 to 4.0, the
with existing ones. In this study, dilation and erosion are experimentation is carried out in the suggested model, the
various characterization calculations are handled with the
used to remove blood vessels. Dilation is as follows;
highlights are removed, and it is discovered that Nave Bayes
Classifier is most effective in comparison to other classifiers
with a precision of 83%. When compared to the present
clustering approaches, the suggested work includes an
extraction strategy that has reduced the complexity of the
existing situation. By using it on low-cost SOCCs like the
Raspberry Pi, this technique may be used to put together an
Where B and B1 are structuring elements and b, b1 are independent, logical critique of machine learning
grids. The combined morphological operation can be stated algorithms. Laplacian of the Gaussian function was used for
as; feature extraction. ConvNet-based calculations were
proposed by Mamta Arora et al. [10] for the detection of
C ′ = (A ⊕ B2) ⊖ B2 − (A ⊕ B1) ⊖ B1 diabetic retinopathy using fundus pictures. This paper
demonstrates the applicability of deep learning as a solution
Another method was put up by KK Palavalasa et al. [6] to this problem. There is still a lot more study to be done in
et al. to identify hard exudates with good accuracy in relation order to improve this model. It has been discovered that the
to the severity of the injury. Using foreground and model can be applied to the best-prepared model scenario,
background techniques, the exudates sores were initially which may significantly improve the results. However, there
differentiated in the current method. After the succeeding is much better trial that will apply in the future to further
steps, the calculation's last progression deleted the loud enhance results. A deep learning-based system was
exudates and adjusted the findings as necessary. The proposed by Yash S. Boral et al. [11]. The system's goal is
computation has been tested in the publically available to categorise the fundus picture and identify DR. This
DirectDB information base, which provides the crucial data. research put out a method for improving picture quality in
With an affectability of 0.87, a F7 score of 0.78, and a order to boost the system's effectiveness and precision. This
positive rating of 0.76, it has proven possible to get higher deep network, known as V3, is in charge of extracting the
execution outcomes for hard exudates painful level finding characteristics of the fundus picture and attempting to
as compared to current tactics. categorise the DR from it. For training purposes, 48
photographs were used, while 90 images were used for

IJISRT23JUL471 www.ijisrt.com 2282

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
testing. SVM has been utilised in the last step of diagnosis
to categorise the condition as either diabetic retinopathy or a
normal picture. A methodology for detecting diabetic
retinopathy using a capsule network was put out by G.
Kalyani [12]. Large datasets cannot be used to test the
capsule network. The inner loop may cause it to slow down
and allow for fewer iterations. Comparing the intricacy of
the two networks, the capsule network is more complicated
than the CNN. Small filters should be used by the network
to make it speedier and take less calculation time.

III. IMPLEMENTATION DETAILS

The proposed method is based on Sobel Edge

Detection and Color Mapping techniques. Color mapping is
a method for transforming RGB images by pixel intensities,
which can be used to draw attention to image flaws. Fig 4 Color Mapping of Input Fundus Image of Messidor
Calibrating a photometric camera is another name for color Benchmark
correction. In the area of image processing, color mapping is
a key activity that aids in the classification of objects. The A. Sobel Edge Detection
fundamental goal of color mapping is to establish a The proposed system is based on Sobel Edge
statistical correlation between an image's color and Detection, whose goal is to identify the edges of an image—
brightness using the image's mean and variance. vertically and horizontally—in order to gather data. Because
Sobel edge detection plays a significant part in the majority
H = {(h(1), v(1)),...,(h(O), v(O))} (1) of contemporary systems, it is a highly well-liked edge
detection technique. The form and density of exudates may
Where H is the histogram, w is the width and O is the be determined with the use of sobel edge detection. It
structure of an image I distinguishes between the main item and background noise.
Edge detection is defined as a quick shift in the brightness
max(𝐼)−min(𝐼) or pixel value of a picture. Prewitt, Canny, Laplacian,
𝑂= [ ]
𝑤 (2) Roberts, and many more edge detection methods exist, but
they are all tailored to the needs of the system. Sobel is
𝑁 regarded as the best edge detection tool available for disease
ℎ(i) = ∑ 𝑃(𝐼(𝑝), i), i ∈ [1, 𝑂] (3) diagnosis.
k=1
 Horizontal Kernel
w(i) = 𝑚i𝑛(𝐼) + (i − 1)w (4)

where i is the pixel intensities and P is the probability

distribution. An method for statistical colour mapping treats
pixel intensities as 3D colour spaces and reshapes the data to
correspond to the statistical points. Various methods map The horizontal kernel mask's direction is leftward,
the colour individually, one channel at a time, depending on which suggests that the calculation should be carried out
the colour chosen. However, the colour mapping method from right to left. And derivative is as follows-
alters many colours at once. The colour mapping of the
fundus picture improves the visibility of exudates and makes
it simple to identify background vessels by concentrating on
damaged cells. The colour mapping of the input fundus
picture for the Messidor benchmark is shown in Fig. 4.

First order derivation is computed in the specified

direction and is sensitive to horizontal edges.

IJISRT23JUL471 www.ijisrt.com 2283

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 Vertical Kernel

When using a vertical kernel mask, the calculation

must be done from right to left while facing upward. First
order derivation is computed in the specified direction and is
sensitive to vertical edges.

Fig 7 Exudates Extraction

Diagnoses based on Sobel edge detection are superior

to those based on machine learning because they can handle
fresh patient data or data that does not pertain to the original
variant. The samples for which a machine learning-based
model has been trained are its only bounds. Therefore, it is
preferable to employ edge detection methods that are
Fig 5 Horizontal & Vertical Coordinates
applicable to all data.
𝑎0 𝑎1 𝑎2
Table 1 Proposed Algorithm
𝑎7 [i, j] 𝑎3
𝑎6 𝑎5 𝑎4

𝑀𝑥 = (𝑎2 + 𝑐𝑎3 + 𝑎4 ) − (𝑎0 + 𝑐𝑎7 + 𝑎6 )

𝑀𝑦 = (𝑎6 + 𝑐𝑎5 + 𝑎4 ) − (𝑎0 + 𝑐𝑎1 + 𝑎2)

Fig 6 Encoding and Decoding Model

B. Exudates Localization
Exudates are the defects in the picture that indicate the
existence of diabetic retinopathy. Exudate differs slightly
from normal cells and blood vessels, but it is difficult to
distinguish them with unaided vision. It may be detected by
medical specialists or by doing a thorough examination of
the fundus picture using a specific approach. The
segmentation of background noise or blood vessels is shown
in Fig. 8, and the system has coloured the exudates in red.

IJISRT23JUL471 www.ijisrt.com 2284

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IV. EXPERIMENTAL RESULT positive (FP). When a system diagnoses normal vision in an
image with DR, it is referred to as a false negative (FN). In
The four parameters True Positive (TP), False Positive (FP), the Messidor benchmark, there are a total of 1212 testing
True Negative (TN), and False Negative (FN) are used to photos, 655 of which include soft and hard exudates, and
base the results of the experiment. If a system correctly 557 of which have none. This indicates that 557 photos are
identified a condition as having either hard or soft exudates from healthy eyes and 655 images are from the class of
on an image, the designation will be TP; otherwise, it will diabetic retinopathy. System identified 550 photos as having
be TN. When a system diagnoses diabetic retinopathy in an healthy symptoms and 642 images as having soft and hard
image with a grade level of 0, it is referred to as a false exudate.

Fig 8 Console Result & GUI

Fig 8 shows the MATLAB console result and Graphical User Interface of proposed system.

IJISRT23JUL471 www.ijisrt.com 2285

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Table 2 Experimental Results

Table 2 shows the result of the proposed system for Messidor benchmark. As per the confusion matrix, the calculations are
as follows;

Table 3 Results Comparison

Table 2 represents the result obtained for all metrics and table III shows the result comparison with capsule network.

V. CONCLUSION & FUTURE SCOPE method. Future research can examine various datasets and
improve accuracy. Future system improvements might
It is difficult for artificial intelligence to diagnose any increase efficiency and accuracy.
ailment from a picture. There are several models that may
be used to perform autonomous diabetic retinopathy REFERENCES
diagnosis, however it is necessary to have greater precision
and faster processing. The proposed approach is based on [1]. Wejdan L. Alyoubi, Wafaa M. Shalash, Maysoon F.
Sobel edge detection and colour mapping. In comparison to Abulkhair, Diabetic retinopathy detection through deep
the machine learning technique, Sysetm can diagnose the learning techniques: A review, Informatics in Medicine
DR more quickly. Exudates may be seen more clearly Unlocked, Volume 20, 2020.
thanks to colour mapping, and Sobel can extract the [2]. Kaggle, APTOS : Eye Preprocessing in Diabetic
exudate's structure and entropy. Comparing the system to Retinopathy,
prior proposed models like the capsule method, it showed https://www.kaggle.com/code/ratthachat/aptos-eye-
remarkable accuracy. Messidor benchmark was held by the preprocessing-in- diabetic-retinopathy/notebook.
system while it reviewed the test results. The accuracy of Accessed 07 April 2022.
the system is 98.35%, somewhat better than the prior

IJISRT23JUL471 www.ijisrt.com 2286

Volume 8, Issue 7, July – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[3]. Towards Data Science, Blindness detection (Diabetic [14]. Salman, Ahmad & Siddiqui, Shoaib & Shafait, Faisal
retinopathy) using Deep learning on Eye retina images, & Mian, Ajmal & Shortis, Mark & Khurshid, Khawar
https://towardsdatascience.com/blindness-detection- & Ulges, Adrian & Schwanecke, Ulrich. (2019).
diabetic- retinopathy-using-deep-learning-on-eye- Automatic fish detection in underwater videos by a
retina-images-baf20fcf409e. Accessed 07 April 2022. deep neural network-based hybrid motion learning
[4]. Prasanna Porwal, Samiksha Pachade, et al. IDRiD: system. ICES Journal of Marine Science. 77.
Diabetic Retinopathy – Segmentation and Grading 10.1093/icesjms/fsz025.
Challenge, Medical Image Analysis, Volume 59, 2020. [15]. P. Kokare, "Wavelet based automatic exudates
[5]. S. Ravishankar, A. Jain and A. Mittal, "Automated detection in diabetic retinopathy," 2017 International
feature extraction for early detection of diabetic Conference on Wireless Communications, Signal
retinopathy in fundus images," 2009 IEEE Conference Processing and Networking (WiSPNET), 2017, pp.
on Computer Vision and Pattern Recognition, Miami, 1022-1025, doi: 10.1109/WiSPNET.2017.8299917.
FL, 2009, pp. 210-217. [16]. N. Karami and H. Rabbani, "A dictionary learning
[6]. K. K. Palavalasa and B. Sambaturu, "Automatic based method for detection of diabetic retinopathy in
Diabetic Retinopathy Detection Using Digital Image color fundus images," 2017 10th Iranian Conference
Processing," 2018 International Conference on on Machine Vision and Image Processing (MVIP),
Communication and Signal Processing (ICCSP), 2017, pp. 119-122, doi:
Chennai, 2018, pp. 0072-0076, doi: 10.1109/IranianMVIP.2017.8342333.
10.1109/ICCSP.2018.8524234. [17]. Dailyhunt, ‘Diabetic retinopathy can cause vision
[7]. A. Osareh, B. Shadgar, and R. Markham, “A loss’, [Accessed: 26-March-2022 [Online]. Available:
computational- intelligence-based approach for https://m.dailyhunt.in/news/india/english/careguru+eng
detection of exudates in diabetic retinopathy images,” lish-epaper-
IEEE Trans. Inf. Technol. Biomed., vol. 13, no. 4, pp. creguru/diabetic+retinopathy+can+cause+vision+loss-
535–545, 2009 newsid- 97367708.
[8]. Gardner, G & Keating, David & Williamson, Tom & [18]. Sisodia D. S, Nair S, Khobragade P. Diabetic Retinal
Elliott, A. (1996). Automatic detection of diabetic Fundus Images: Preprocessing and Feature Extraction
retinopathy using an artificial neural network: A for Early Detection of Diabetic Retinopathy. Biomed
screening tool. The British journal of ophthalmology. Pharmacol J 2017.
80. 940-4. 10.1136/bjo.80.11.940. [19]. Klein R, Klein BE, Moss SE, Davis MD and DeMets
[9]. M. M. Dharmana and A. M.S., "Pre-diagnosis of DL, “The Wisconsin epidemiologic study of diabetic
Diabetic Retinopathy using Blob Detection," 2020 retinopathy. II Prevalence and risk of diabetic
Second International Conference on Inventive retinopathy when age at diagnosis is less than 30
Research in Computing Applications (ICIRCA), years,” Arch Ophthalmology 1984, vol. 102, pp. 527–
2020, pp. 98-101, doi: 532.
10.1109/ICIRCA48905.2020.9183241. [20]. B. Harangi, I. Lazar and A. Hajdu, “Automatic
[10]. M. Arora and M. Pandey, "Deep Neural Network for Exudate Detection Using Active Contour Model and
Diabetic Retinopathy Detection," 2019 International Region wise Classification,” IEEE EMBS 2012,
Conference on Machine Learning, Big Data, Cloud and pp.5951–5954.
Parallel Computing (COMITCon), 2019, pp. 189-193, [21]. Balazs Harangi, Balint Antal and Andras Hajdu,
doi: 10.1109/COMITCon.2019.8862217. “Automatic Exudate Detection with Improved Nave-
[11]. Y. S. Boral and S. S. Thorat, "Classification of Bayes Classifier, Computer- Based Medical Systems,”
Diabetic Retinopathy based on Hybrid Neural CBMS 2012, pp. 1–4.
Network," 2021 5th International Conference on [22]. K Zuiderveld, “Contrast Limited Adaptive Histogram
Computing Methodologies and Communication Equalization,” Graphics Gems IV, Academic Press
(ICCMC), 2021, pp. 1354-1358, doi: 1994, pp. 474–485.
10.1109/ICCMC51019.2021.9418224. [23]. M. N. Langroudi and Hamed Sadjedi, “A New Method
[12]. Kalyani, G., Janakiramaiah, B., Karuna, A. et al. for Automatic Detection and Diagnosis of Retinopathy
Diabetic retinopathy detection and classification using Diseases in Colour Fundus Images Based on
capsule networks. Complex Intell. Syst. (2021). Morphology,” International Conference on
https://doi.org/10.1007/s40747-021- 00318-9 Bioinformatics and Biomedical Technology 2010, pp.
[13]. Kanimozhi, J., Vasuki, P. & Roomi, S.M.M. Fundus 134–138.
image lesion detection algorithm for diabetic [24]. S. Chaudhauri, S. Chatterjee, N. Katz, M. Nelson and
retinopathy screening. J Ambient Intell Human M. Goldbaum, "Detection of blood vessels in retinal
Comput 12, 7407–7416 (2021). images using two dimensional matched filters," IEEE
https://doi.org/10.1007/s12652-020-02417-w Trnas. Medical imaging, vol. 8.
[25]. X. Jiang and D. Mojon, “Adaptive local thresholding
by verificationbased multithreshold probing with
application to vessel detection in retinal images,” IEEE
Trans. Pattern Anal. Mach. Intell., vol. 25, no. 1,pp.
131–137, Jan. 2003.

IJISRT23JUL471 www.ijisrt.com 2287