International Conference on Artificial Intelligence and Machine Learning

Diabetic Retinopathy Detection using Image Processing (GUI)

T.N.R.Nithina, U.M.S.Koushikb, Dr.R.Subhashinic
a.b
Student, Computer Science and Engineering, Sathyabama Institute of Science and Technology, Chennai 600119, India
c
Professor, School of Computing, Sathyabama Institute of Science and Technology, Chennai 600119, India

Abstract

Diabetic Retinopathy is a major disease that has affected over 290 million people globally and 69.2 million people in India, the
rate of people getting affected will increase exponentially in the coming years. Diabetic Retinopathy is an ailment linked to the
fundus of the eye and can have adverse effects on the patient, if at all left undiagnosed respectively. Our project aims to construct
a graphical user interface that can integrate image processing techniques together in order to predict whether the input
fundus/retinal image received from the patient is affected with Diabetic Retinopathy or not; if affected, the graphical user
interface will display the severity along with the required action needed to be undertaken by the user / patient. This essentially
reduces the processing time involved in the process of detecting the disease and also the ophthalmologists can also have our
graphical user interface as a backup that can be used for validating or assist in detecting the disease.

Keywords: Diabetic retinopathy; Graphical user interface; Image processing; Fundus analysis; Retinal ailment detection.

1. Introduction

Diabetic retinopathy is an infection that happens mostly in working populace for the most part experiencing
diabetes. It for the most part includes a great deal of time in preparing the pictures of the fundus after a patient visits
the ophthalmologist separately. Our undertaking plans to lessen that with the goal that specialist can take care of
more patients in light of the quicker handling of results separately and furthermore to guarantee that there is no
misdiagnosis occurring, by helping the ophthalmologist. There are 4 arranges in Diabetic retinopathy which are to
be specific, Mild Diabetic Retinopathy, Moderate Diabetic Retinopathy, Severe Diabetic Retinopathy, Proliferative
Diabetic Retinopathy separately. Proceeding with carelessness of not visiting an authority or not taking
consideration can result doubtlessly in visual deficiency, making the patient be visually impaired for the remainder
of their lives. The three fundamental segments of the human retina are vessels, optic circle and fovea. what's more,
are for the most part utilized for a few applications, For instance, retinal picture selection, lighting up revision,
similarly as pathology acknowledgment interior of the retina. Distinguishing proof for these fundamental structures
physically is monotonous and depends upon the bent of the customer. The division of veins from fundus photos can
be troublesome for various reasons. A portion of the tainting

c Dr.R.Subhashini. Tel. : +914424503160

E-mail address: [email protected]
sources are identified with the securing procedure and sort of symbolism, also the others are inborn highlights of
fundus pictures. The powerful factors that make the division troublesome must be the inappropriate retinal picture
differentiate and also the rough foundation light. The bumpy light is because of the securing procedure, and the
unacceptable complexity is due to the procurement procedure and also the process by which that diverse vessels has
distinctive stand out from foundation. Accordingly, nearness of a programmed vein identification instrument that
sections the veins of retina in a brief span and with high precision is our craving. Numerous endeavors have been
made and different techniques have been acquainted all together with fragment retinal pictures. The calculations in
this field fall in the following gatherings: window-based, classifier-based, and following based methodologies. The
Window-based techniques, for example, edge discovery, gauge a match at every pixel for a given model against the
pixel's encompassing window. The cross segment of a vessel in a retinal picture was demonstrated by a Gaussian-
molded bend and after that recognized utilizing pivoted coordinated channels. The vertebrate retina is a light touchy
tissue covering the internal surface of the eye. The optics of the eye makes a picture of the visual world on the
retina, which serves much a similar capacity as the film in a camera. Light striking the retina starts a course of
synthetic and electrical occasions that at last trigger nerve driving forces. These are sent to different visual focuses
of the mind through the strands of the optic nerve. At the point when brilliant light is shone on the eye, it will
consequently contract. This is the drag scriptural reflex, which is an essential trial of brainstem work. Besides, the
student will enlarge if an individual sees an object of intrigue. The oculomotor nerve, explicitly the parasympathetic
part originating from the Edinger-Westphal core, ends on the roundabout iris sphincter muscle. At the point when
this muscle contracts, it lessens the span of the understudy. A standout amongst the most terrifying entanglements of
diabetes is the degenerative eye malady diabetic retinopathy. A conclusion of this condition must be made by an
accomplished ophthalmologist, despite the fact that there are a few signs and side effects of diabetic retinopathy that
patients ought to know about. while all diabetes patients are in danger of creating diabetic retinopathy, not every one
of them are bound for visual deficiency. There are a few critical signs and side effects to search for with regards to
diabetic retinopathy. Albeit a significant number of these side effects are absent in the soonest phases of non-
proliferative diabetic retinopathy, they frequently present themselves as the infection advances toward cutting edge
or proliferative diabetic retinopathy. In the event that you have type I or II diabetes, you are in danger for building
up this degenerative eye ailment. One has to get familiar with the reasons for diabetic retinopathy and how it can
prompt vision misfortune. Numerous diabetic patients are stunned when they get a diabetic retinopathy analysis and
discover that they've been living with a degenerative eye ailment and never took note. Frequently it's not until
patients start to lose their ordinary vision that they understand they have to see a specialist. The accompanying signs
and side effects of diabetic retinopathy are ordinarily displayed after the malady has officially compounded, similar
to fogginess, obscuring of the retina, tingling and swelling of the eye inside, torment in the eye once in a while for
much of the time as well. On the off chance that you have encountered any of these side effects, it is emphatically
prescribed that you contact an accomplished ophthalmologist as quickly as time permits. Likewise, a few side
effects of diabetic retinopathy may change amid the movement of the infection. Despite the fact that a patient may
have no vision disability, retinal changes saw amid a dream test can be indications of diabetic retinopathy. These
progressions incorporate retinal swelling, veins spilling liquids, and any scar tissue or strange stores on the retina.
Therefore, it is the on the patient’s side as well to ensure the disease does not develop into a higher stage, which
might have very bad effects on the patient. Our project also will reduce the number of reviews required by the
ophthalmologists, which essentially means, our project will be useful in reducing the time spent in analyzing the
patient’s condition.

2. Related Work

As advancements have taken place, a ton of studies and research have occurred on programmed determination of
diabetic retinopathy utilizing various progressed and increasingly exact exactness procedures. The following papers
were studied and deeply analyzed in order to develop our project. The papers which are mentioned below are
published as a part of renowned journals and were truly helpful in understanding the idea of Diabetic Retinopathy.
2.1 Diabetic Retinopathy Detection using Deep Convolutional Neural Networks
Authors: Darshit Doshi, Aniket Shenoy, Deep Sidhpura, Dr. Prachi Gharpure
Year: 2016

Diabetic retinopathy is when harm strikes the retina because of diabetes, which influences up to 80 percent of all
patients who have had diabetes for a long time or more. The mastery and hardware required are frequently ailing in
regions where diabetic retinopathy discovery is generally required. The vast majority of the work in the field of
diabetic retinopathy has been founded on ailment identification or manual extraction of highlights, yet this paper
goes for programmed conclusion of the ailment into its diverse stages utilizing profound learning. This paper
displays the plan and execution of GPU quickened deep convolutional neural networks to naturally analyze and
subsequently group high-goals retinal pictures into 5 phases of the infection dependent on seriousness. The single
model precision of the convolutional neural networks exhibited in this paper is 0.386 on a quadratic weighted kappa
metric and ensembling of three such comparative models brought about a score of 0.3996.

2.2 Diabetic retinal exudates detection using machine learning techniques

Authors: P.R.Asha, S.Karpagavalli
Year: 2015

Diabetic Retinopathy (DR) is an eye filled sickness brought about by the entanglement of polygenic ailment and
that will be distinguished precisely for convenient treatment. As polygenic ailment advances, the vision of a patient
could start to fall apart and prompts visual impairment. In this proposed work, the nearness or nonattendance of
retinal exudates are identified utilizing Machine Learning strategies. To recognize the nearness of exudates
highlights like Mean, Standard deviation, Centroid and Edge Strength are removed from Luv color space in the
wake of sectioning the Retinal picture. A sum of 100 pictures were utilized, out of which 80 pictures were utilized
for preparing / training and 20 pictures were utilized for testing. The grouping task completed with classifiers like
Naive bayes (NB), Multilayer Perceptron (MLP) and Extreme Learning Machine (ELM). Trial results demonstrates
that the model fabricated utilizing Extreme Learning Machine outflanks other two models and adequately
distinguishes the presence of exudates in retinal pictures.

2.3 Automatic Detection of Diabetic Retinopathy using Image Processing and Data Mining Techniques
Authors: Ketki S. Argade, Kshitija A. Deshmukh, Madhura M. Narkhede, Nayan N. Sonawane, Sandeep Jore
Year: 2015

Diabetic Retinopathy is an ailment of retina which influences patients with diabetes mellitus and it is a primary
purpose behind visual impairment. It is an illness in which the retinal veins swell. This harms the retina of eye and
may prompt visual impairment if the severity of diabetes is extremely high. The best treatment is early identification
through normal screenings. Programmed screening of these pictures would push the specialists to effortlessly
distinguish the patient's condition in increasingly precise way. This stresses on assurance of retinal pictures utilizing
suitable picture handling and information mining procedures. With this we can without much of a stretch order
ordinary and irregular pictures of retina this will decrease the quantity of surveys for the ophthalmologist.

3. Proposed system

The propounded design deals with various image processing techniques and machine learning methods so as to
segment the fundus input image properly, and are explained in the upcoming subsection respectively. Our system
indeed has quite a few advantages like; very intelligent and the Graphical User Interface has an effectively
recognizable interface, which could be utilized by anybody. Also, it involves generally faster processing time than
similar systems. The system is cost viable as this can cost lesser than the existing manual strategies. The look and
non intricate feel has better ease of use when contrasted with the past prescient models. More solid, as the
preparation was finished utilizing a dependable informational collection comprising of pictures taken
straightforwardly from a dataset that was built by ophthalmologists individually (DIATREB). The biggest advantage
is of course to assist the ophthalmologist in reviewing the condition of the patient. However, there are a few
drawbacks also to be considered, like; systems created were not at all interactive and perceivable to common non –
technical people. Unlike the reliable fundus images, mobile captured images can have limitations involved with
them like lack of clarity that may lead to misclassification.

3.1 Architecture

Fig. 1. Architecture Diagram

The basic input of the proposed system will be a fundus image that is produced after screening tests respectively.
A fundus is a inner part of retina. This is given by the user using a Graphical User Interface (GUI). The user should
open the interface and then click on the upload image button respectively to upload the image. This Image then
undergoes image pre-processing. This involves changing the images into the required or effective form for
processing. Firstly, the images are resized and sampled into the required size, say 512 X 512 if required. Then color
space conversions and filtering techniques like Gaussian Filtering, Bilateral Filtering, Grey Scale Conversion are
employed to convert the image into gray scale, reduced noise image respectively. After which edge detection
methods and contouring techniques are used to outline the image properly. The method Gaussian blurring is utilized
to reduce the presence of noise in the image and segmentation is used for detecting the area where the blood vessels
are present. Finally, the concerned area is identified via a method called K-means image segmentation and is used
for determining the GLCM (Gray level Co-occurrence Matrix) features matrix of the image. After comparing the
different metrics of the input image’s glcm features, with all of those in the training data set, the model comes to a
decision by judging the image with the severity of DR based on the training image features.

3.2 Algorithms Used

i. Gaussian filtering:
Gaussian filtering is a method which can be used for blurring out the input images which helps in removal of any
presence of noise in the image. The Gaussian filter is not a uniform low pass filter. When we work with images we
usually use a two-dimensional version of gaussian function. A sample Gaussian filter is shown below

Fig.2. Gaussian filter implementation

ii. Bilateral Filtering:

A bilateral filtering is a non-straight, edge-safeguarding, and commotion diminishing smoothing channel for
pictures. It replaces the power of every pixel with a weighted normal of force esteems from close-by pixels. This
weight can be founded on a Gaussian conveyance. Significantly, the loads depend on Euclidean separation of pixels,
yet in addition on the radiometric contrasts (e.g., range differences, such as color intensity, depth distance, etc.). This
preserves sharp edges See Fig[2].

Fig.3. A bilateral filtered image

iii. Colour space conversion (Edge detection):

Transformation of a shading picture into a grayscale picture comprehensive of remarkable highlights is a

convoluted procedure. The strategy of advancement over greyscale picture may subsequently lose contrasts,
sharpness, shadow, and structure of a shading picture. To defend all of the before mentioned attributes of the
shading picture, another count has proposed. To change over the shading picture into grayscale picture the new
calculation undertakes RGB supposition, lessening, and extension of chrominance and luminance. The grayscale
pictures made using the computation in the test asserts that the figuring has ensured the remarkable highlights
belonging to the shading picture, for example sharpness, contrasts, picture structure and shadow.

Fig.4. A grayscale converted image

iv. K means Clustering and Segmentation:

After the entire process is done, the K means technique is utilized to determine the interest region from the feed in
image depending on pixel level intensity respectively. Currently the classifier classifies based on the pixel level
produced
from the feature matrix with each other image in the data set respectively. Here K-means approach is utilized to
get the region of interest needed for future steps to be done. The below image describes k-means clustering .

Fig.5. K-means clustering

4 Methodology

The methodology follows a simpler approach than the existing once. The user interface enables the person to
provide a fundus image of the eye. This is then sent to the model. The model initially apply the gaussian blurring.
This removes the presence of noise in the images to be processed. Later the image is bilateral filtered and
contouring is done. After this is done the colour space conversion takes place where we create a grayscale image of
fundus. Now we apply k-means clustering for identifying the region of interest.
At that point the division happens. The ROI that is gotten is then prepared to remove the GLCM highlights. It
includes the element vector for grouping which comprises of seven highlights acquired by means of division of
retinal structures and surface investigation on entropy , differentiate , homogeneity. The beneath picture is a preview
of the preparation procedure.

This picks the grayscale power of each pixel present as a structure. These characteristics are then used to get
estimations. This is moreover completed with the accessibility dataset pictures. The section extraction of these
photos is done and the estimations of both planned dataset and the customer input are showed up differently in
association with pick the segment of validity the data picture has. Subordinate upon the validity a particular kind of
proposition will be given to the customer.

Currently, we are attempting to test the retinal ailment along these lines utilizing the impelled picture dealing
with strategies on pictures of the retinal. The most basic changes that can develop in the fundus in perspective of the
ailment are Exudates, Haemorrhages and Micro aneurysms which are not discussed widely much in the paper.
Nearness of something like significant quantity of the above wounds in the fundus exhibits closeness to DR.

The acknowledgment techniques join the two which are, managed and non-directed frameworks, non-supervised
strategy is being employed here for identifying hard exudate. Picture Pre-processing incorporates changing the
photos into the required or practical structure for dealing with. That is the reason it is named as pre dealing with.
Directly off the bat, the photos are resized and inspected into the required size, say 512 X 512 at whatever point
required. By then shading space changes like green scale change hop out at devotee the image into dim scale.

After which edge recognizable proof procedures are used to chart the image properly. Centre purifying is utilized
for reducing the upheaval existing in the image and division is used to distinguish the veins evidently. Feature
extraction, It basically incorporates the component vector used for gathering includes seven features got from
division of retinal structures and surface analysis. Classifier, after the whole taking care of is done, the convolutional
neural framework is used to organize the image by means of setting up the image with its several layers in order to
check if the portrayal is immaculately done by the model which is the best way to set up images differentially.
 This stage finally returns irrespective of the ailment still being found in the person suffering. Graphical user
interface, This is the front end obviously, where the user is asked to upload the fundus image by clicking the button
provided by the GUI respectively. It is not completely interactive though. The chat application interface is designed
to facilitate easier navigation for the users who are not very technically qualified.

Displaying the outcomes is when finally after all of the processing is done, the graphical user interface displays
textually whether the person is affected with diabetic retinopathy or not. If at all affected, the bot suggests the
symptoms also regarding the severity of the Diabetic Retinopathy is finally detected and displayed on to the GUI
respectively so that the patient can know about the condition of his ailment currently faced by the potential patient.

5. RESULTS AND DISCUSSION

The outcomes that are acquired throughout the process are as listed below respectively with the captions and
the initial and final phases of the fundus image input by the patient respectively.

i. Initial fundus image: ii. Gaussian filtered fundus image:

Fig.6 Initial fundus image Fig.7 Gaussian filter image

iii. Bilateral filtered fundus image: iv. Grayscale converted fundus image:

Fig.8 Bilateral filter image Fig.9 Grayscale converted image

 v. Segmented fundus image: vi. Feature extraction : GLCM features

Fig.11 Segmented fundus image Fig.12 Snapshot of GLCM matrices

vii. Graphical User Interface: viii. Result screenshot:

Fig.13 Screenshot of GUI Fig.14 Screenshot of Result

ix. Types of DR present:

Fig. 15. (a) normal fundus; (b)fundus with mild-DR; (c) fundus with moderate-DR; (d) fundus with severe-DR;(e) fundus with proliferative DR
6. Conclusion and Future Work

Currently in the scope of this paper, the processing time of diabetic retinopathy detection is kept in mind, so as
to deliver a graphical user interface to the user / patient, so that they can use their fundus images as input to the
graphical user interface and there by get to know whether they are suffering from diabetic retinopathy or not from a
reliable source with faster processing time also. Further upgrades later on might incorporate the use of a superior
calculation something like a convolutional neural system which can help in arranging the pictures well than the
present utilized classifier individually. Aside from that, highlights like helpline or client manual ought to be given
in the graphical UI, to be useful for the clients who probably won't be acquainted with innovative progressions and
utilization of the application.

Acknowledgement

We wish to acknowledge the Department of Science and Technology, India and School of Computing, Sathyabama
Institute of Science and Technology, Chennai for providing the facilities to do the research under the DST-FIST
Grant Project No.SR/FST/ETI-364/2014.

References

1. M. E. Martinez-Perez, A. D. Hughes, S. A. Thom, and K. H. Parker, “Improvement of a retinal blood vessel

segmentation method using the insight segmentation and registration toolkit (ITK),” in Proc. IEEE 29th Annu.
Int. Conf. EMBS. Lyon, IA, France, 2017, pp. 892– 895.
2. S. Dua, N. Kandiraju, and H.W. Thompson, “Design and implementation of a unique blood-vessel detection
algorithm towards early diagnosis of diabetic retinopathy,” in Proc. IEEE Int. Conf. in Inf. Technol., Coding
Comput., 2015.
3. Darshit Doshi, Aniket Shenoy, Deep Sidhpure, Prachi Gharpure, “Diabetic Reinopathy Detection using
Deep Neural Networks”. 2016 IEEE Confeernce.
4. A.M.Mendonca and A. Campilho, “Segmentation of retinal blood vessels by combining the detection of centerlines
and morphological reconstruction,”IEEE Trans. Med. Imag., vol. 25, no. 9, pp. 1200–1213, Sep. 2014.
5. A. Aquino, M. E. Gegúndez-Arias, and D. Marín, “Detecting the optic disc boundary in digital fundus images
using morphological, edge detection, and feature extraction techniques,” IEEE Trans. Med. Imag., vol. 29, no.
11, pp. 1860–1869, Nov. 2012.
6. Klein R, Klein BE, Moss SE, Davis MD and DeMets DL, “The Wisconsin epidemiologic study of diabetic
retinopathy. II Prevalence and risk of diabetic retinopathy when age at diagnosis is less than 30 years,” Arch
Ophthalmology 2014, vol. 102, pp. 527–532.
7. B. Harangi, I. Lazar and A. Hajdu, “Automatic Exudate Detection Using Active Contour Model a nd Region
wise Classification,” IEEE EMBS 2015, pp.5951–5954.
8. Deepthi K Prasad, L.Vibha, K.R.VenuGopal, “Early detection of diabetic retinopathy from digital retinal
fundus images” , 2015 IEEE Conference.
9. Kanika Verma, Prakash Deep, A.G Ramakrishnan, “Detection and classification of diabetic retinopathy using
retinal images”. 2011 IEEE