Visvesvaraya Technological University

Jnana Sangama, Belagavi - 590018

A Project Work Phase - 2 (18CSP83)

Report on
“AUTOMATIC DETECTION AND CLASSIFICATION OF
DIABETIC EYE DISORDERS”
Project Report submitted in partial fulfillment of the requirements for the award of
the degree of
BACHELOR OF ENGINEERING
IN
COMPUTER SCIENCE AND ENGINEERING
Submitted by
Deeksha B 1KG20CS024
Charishma C 1KG20CS019
Jyothsna B 1KG20CS014
Bhoomika T 1KG20CS012

Under the Guidance of

Mrs. NITA MESHRAM
Associate Professor,
Department of Computer Science & Engineering KSSEM,
Bengaluru - 560109

K. S. School of Engineering and Management

DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING
#15,Mallasandra, off. Kanakapura Road, Bengaluru – 560109
2023 – 2024
 K. S. School of Engineering and Management
#15, Mallasandra, Off, Kanakapura Road, Bengaluru - 560109

Department of Computer Science & Engineering

CERTIFICATE

Certified that the Project Work Phase-II (18CSP83) entitled “Automatic detection and classification
of diabetic eye disorders” is a bonafide work carried
out by:
Deeksha B 1KG20CS024
Charishma C 1KG20CS019
Jyothsna B 1KG20CS014
Bhoomika T 1KG20CS012

in partial fulfillment for VIII semester B.E, Project Work in the branch of Computer Science and
Engineering prescribed by Visvesvaraya Technological University, Belagavi during the period of
February 2024 to May 2024. It is certified that all the corrections and suggestions indicated for
internal assessment have been incorporated. The Project Work Phase-II Report has been approved as it
satisfies the academic requirements in report of project work prescribed for the Bachelor of
Engineering degree.

….…………………………………………… ….…………………………………………… ….……………………………………………

Signature of the Guide Signature of the HOD Signature of the Principal
[ Mrs. Nita Meshram] [ Dr. K Venkata Rao] [ Dr. K Rama Narasimha]
Associate Professor, CSE Professor & Head, CSE Principal/ Director
K.S.S.E.M, Bengaluru K.S.S.E.M, Bengaluru K.S.S.E.M, Bengaluru
 DECLARATION

We, the undersigned students of 8th semester, Computer Science & Engineering, KSSEM,
declare that our Project Work Phase-II entitled “Automatic detection and classification of
diabetic eye disorders”, is a bonafide work of ours. Our project is neither a copy nor by
means a modification of any other engineering project.

We also declare that this project was not entitled for submission to any other university in the
past and shall remain the only submission made and will not be submitted by us to any other
university in the future.

Place:

Date:

Name and USN: Signature

Deeksha B (1KG20CS024)

Charishma C (1KG20CS019)

Boyapati Jyothsna (1KG20CS014)

Bhoomika T (1KG20CS012)

I
 ACKNOWLEDGEMENT
The satisfaction and euphoria that accompany the successful completion of any task will be
incomplete without the mention of the individuals, we are greatly indebted to, who through
guidance and providing facilities have served as a beacon of light and crowned our efforts with
success.

We would like to express our gratitude to our MANAGEMENT, K.S. School of Engineering
and Management, Bengaluru, for providing a very good infrastructure and all the kindness
forwarded to us in carrying out this project work in college.

We would like to express our gratitude to Dr. K.V.A Balaji, CEO, K.S. School of Engineering
and Management, Bengaluru, for his valuable guidance.

We would like to express our gratitude to Dr. K. Rama Narasimha, Principal, K.S. School of
Engineering and Management, Bengaluru, for his valuable guidance.

We like to extend our gratitude to Dr. K Venkata Rao, Professor and Head, Department of
Computer Science & Engineering, for providing a very good facilities and all the support
forwarded to us in carrying out this Project Work Phase-II successfully.

We also like to thank our Project Coordinators, Mrs. Supriya Suresh Suchindra, Asst.
Professor, Mrs. Meena G, Asst. Professor, Department of Computer Science &
Engineering for their help and support provided to carry out the Project Work Phase-II
successfully.

Also, we are thankful to Mrs. Nita Meshram, Associate Professor for being our Project
Guide, under whose able guidance this project work has been carried out Project Work Phase-II
successfully.

We are also thankful to the teaching and non-teaching staff of Computer Science &
Engineering, KSSEM for helping us in completing the Project Work Phase-II work.
DEEKSHA B 1KG20CS024
CHARISHMA C 1KG20CS019
JYOTHSNA B 1KG20CS014
BHOOMIKA T 1KG20CS012

II
 ABSTRACT

Eye disease detection using deep learning has emerged as a groundbreaking approach in the
field of medical diagnostics. This technology leverages advanced neural networks to analyze
various ocular conditions with remarkable accuracy and speed. By processing high- resolution
medical images of the eye, deep-learning models can identify and classify diseases such as
glaucoma, diabetic retinopathy, and macular degeneration. The utilization of convolutional
neural networks (CNNs) and recurrent neural networks (RNNs) in tandem with large
annotated datasets has significantly improved detection rates and reduced false positives.
Furthermore, the integration of real-time image analysis and telemedicine has expanded the
reach of eye disease diagnosis, enabling remote monitoring and timely intervention. However,
challenges remain, including the need for extensive and diverse datasets, the interpretability of
deep learning models, and regulatory compliance. Nevertheless, the promising results
achieved thus far underscore the potential of deep learning in revolutionizing eye disease
detection, ultimately enhancing patient outcomes and reducing the global burden of
preventable vision impairment. This paper explores the current state of the field, discusses
challenges and future directions, and highlights the transformative impact of deep learning in
the domain of eye disease diagnosis.

III
 TABLE OF CONTENTS

Chapter Contents Page No.

No.

DECLARATION I

ACKNOWLEDGEMENT II

ABSTRACT III

TABLE OF CONTENTS IV

LIST OF FIGURES VII

Chapter 1 INTRODUCTION 1-2

1.1 OVERVIEW 1

1.2 PURPOSE OF THE PROJECT 1

1.3 SCOPE OF THE PROJECT 2

1.4 DEFINITIONS 2

Chapter 2 LITERATURE SURVEY 4-6

2.1 MULTI DISEASE DETECTION IN RETINAL IMAGES USING

DEEP NEURAL NETWORK 4

2.2 ENSEMBLE LEARNING FOR MULTIPLE EYE DISEASE

CLASSIFICATION 4

2.3 A TRANSFER LEARNING APPROACH FOR MULTI DISEASE

EYE DIAGNOSIS 4

2.4 FUSION OF MODALITIES FOR MULTI DISEASE EYE

DIAGNOSIS 5

2.5 ADVERSARIAL TRAINING FOR ROBUST MULTI EYE

5
DISEASE DETECTION

2.6 REAL TIME MULTI DISEASE DETECTION USING LIGHT

5
WEIGHT NETWORK

IV
 2.7 A COMPREHENSIVE REVIEW OF TRANSFER LEARNING IN
MEDICAL IMAGING FOR EYE DISEASE DETECTION 6

2.8 NOVEL STRATEGY IN DEEP TRANSFER LEARNING FOR

6
PUBLIC HEALTH SYSTEM

2.9 EVALUATION METRICS IN TRANSFER LEARNING FOR

6
EYE DISEASE DETECTION: A COMPARATIVE ANALYSIS

2.10 COST EFFECTIVE FUNDUS IMAGING: SURVEY OF LOW-

COST EQUIPMENT 6

Chapter 3 PROBLEM IDENTIFICATION 7

3.1 PROBLEM STATEMENT 7

3.2 PROJECT SCOPE 7

Chapter 4 GOALS AND OBJECTIVES 8

4.1 PROJECT GOALS 8

4.2 PROJECT OBJECTIVES 8

Chapter 5 SYSTEM REQUIREMENTS SPECIFICATION 9-10

5.1 SOFTWARE REQUIREMENT ANALYSIS 9

5.2 HARDWARE REQUIREMENT ANALYSIS 10

Chapter 6 METHODOLOGY 11-15

6.1 WORKING FLOW OF THE APPLICATION 11

6.2 DIAGRAM OF DATA FLOW 12

6.3 METHODOLOGIES 12

V
Chapter 7 IMPLEMENTATION 16

7.1 FILES USED 16

7.2 MODULES AND THEIR ROLES 16

7.3 CODE FOR DISEASE DETECTION 22

Chapter 8 TESTING 29

8.1 TESTING PRINCIPLE 29

8.2 TESTING METHODS 29

Chapter 9 RESULTS AND SNAPSHOTS 31-33

Chapter 10 APPLICATIONS 34

Chapter 11 CONTRIBUTION TO SOCIETY 35

AND ENVIRONMENT

REFERENCES 36

APPENDIX 1 37

APPENDIX 2 40

APPENDIX 3 46

VI
 LIST OF FIGURES

Fig. No. Figure Name Page

No.
6.1 DESIGN OF THE APPLICATION 11

6.2 DATA FLOW DIAGRAM OF LEVEL 0 12

6.3 DATA FLOW DIAGRAM OF LEVEL 1 12

6.3.1 DATA PREPROCESSING 13

6.3.2 DATA PREPARATION 13

6.3.3 MODEL TRAINING 14

6.4.4 SAVED MODEL 15

8.1 ACCURACY PREDICTION 31

8.2 HOME PAGE 31

8.3 USER REGISTRATION 32

8.4 IMAGE PROCESSING 32

8.5 FINAL OUTPUT 33

VII
 Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
Chapter 1
INTRODUCTION

1.1 Overview

India life care systems have been focusing on artificial intelligence technologies for prompt
diagnosis. Health data must still be reported in a consistent format for machine learning to account
for various features and become more accurate and reliable. Patient condition was treated using a
range of ML approaches, using Decision Trees, Naive Bayes, and Neural Network algorithms,
depending on characteristics like age, medical history, and clinical observations. The suggested
system may identify and detect eye illnesses through source of data mining and retinal scan
techniques.

The purpose of this research is to use AI to check for various eye conditions. has given a summary
of the improvements made in the use of AI and DL technologies for the problem identification of eye
diseases, along with the challenges that DL implementation in screening programmes is now
encountering and the translation of DL research into useful clinical screening applications in a
community environment. A consequence of retinopathy known as diabetic macular edoema occurs
when fluid builds up in the macula, impairing central vision. Diabetes also raises the risk of
glaucoma, which harms the optic nerve, and cataracts, which cause the lens of the eye to cloud. It is
essential to get routine eye exams to find out and treat these conditions early on. Retaining a healthy
lifestyle and maintaining ideal blood pressure and blood glucose levels are high for the prevention
and treatment of diabetic eye disorders.

1.2 Purpose of the Project

Every human being depends on their eyes to see things around them, keeping them an
indispensable component of existence. Sight is most valuable senses since it provides us with 80%
of the information we take in. By taking good care of our eyes, we may prevent blindness and loss
of vision and monitor the have emergence of disorders like cataracts, glaucoma, and diabetic
retinopathy. Most of the people will have eye problems. While minor eye conditions can be treated
easily at home and resolve on their own, more critical situations require medical attention from
qualified professionals. When these ocular conditions are correctly identified at an at an early stage,
only then the progression of these eye diseases can be stopped. These eye diseases have a wide range
of visually to find out symptoms.

Dept. of CSE, KSSEM 1

 Automatic Detection and classification of Diabetic Eye Disorders 2023-2024

1.3 Scope of the Project

Diabetic eye disorders refer to a group of retina problem that causes individuals with diabetes.
These problems are formed by changes in the blood vessels of the retina, the light-sensitive tissue at
the eye nerve. The most common diabetic eye disorders. The scope of diabetic eye disorders
encompasses various facts, including epidemiology, healthcare infrastructure, research, technology,
public health, and global collaboration.

1.4 Definition

1.4.1 Machine Learning

ML is a subfield of AI, which is broadly defined as the scope of a machine to imitate intelligent
human behavior. Uses data and algorithms to enable AI to imitate the way that human learn, gradually
improving its accuracy. There are many different uses for machine learning of algorithms.

1.4.2 Random Forest Algorithm

The classifier enhances the predicting accuracy of datasets by utilizing multiple decision trees on
different subsets and averaging the results. An algorithm is a type of supervised learning method. In
ML, it can be used to both regression and classification issues.

1.4.3 Decision Tree

A decision tree is of a structure that assembles a tree in which of the each internal node evaluates an
attribute, each branch denotes the value of an attribute, and each leaf node shows the conclusion or
prediction. Regression and the classification are two applications for decision trees, a non-parametric
supervised learning technique. The decision tree which is depicted with its root at the top and turned
upside down. In the figure on the left, each bolded word in black stands for an internal node or
condition that determines how the tree divides into branches and edges. The decision or leaf, which
in this case indicates whether the passenger survived or perished, is the end of the branch that doesn't
divide any more. It is shown as red and green letters, respectively

Dept. of CSE, KSSEM 2

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024

1.4.4 Data Cleaning

Data cleaning is a crucial step in the branch of machine learning pipeline, as it involves identifying
and removing any missing, duplicate, or irrelevant data. Raw data may contain numerous errors,
which can change the speed of ML models and leads to incorrect predictions and negative business
impact. Data cleaning will be used out in bulk using scripting or a data quality firewall, or
interactively using data wrangling tools.

1.4.5 Data Preprocessing

Preparing raw data to do it appropriate for a ML model called as data preprocessing. One method
for transforming the raw data into a clean data set is called as the "data preprocessing." To put it
another way, when data is acquired in raw format from multiple sources, it is impractical for
analysis. need to lower the complexity of machine learning algorithms

1.4.6 Python

The programming language Python is well-liked. It was made available. Is an object-oriented,

interpreted, high-level language with dynamic semantics. It is particularly appealing for
developing applications quickly and for usage as a scripting or glue language to join existing
components together because of its high-level built-in data structures, dynamic typing, and
dynamic binding. Python promotes code reuse and software modularity by supporting modules
and packages. For free on all major systems, both the Python interpreter and the large standard
library are accessible in source or binary format. They may be shared without restriction.

Dept. of CSE, KSSEM 3

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
Chapter 2

LITERATURE SURVEY

2.1 Title: “Ensemble Learning for Multiple Eye Disease Classification”

In 2019, the World Health Organization stated that 2.2 billion people worldwide have a visual
impairment or blindness, half of which were either preventable or were not yet addressed. 1 The
limited number of eye health professionals, especially in certain populations and geographic areas,
is a barrier to more widespread in-person eye screening. One way to address this gap in coverage is
through detection of eye pathologies using artificial intelligence (AI). This allows for efficient remote
screening followed by prompt patient referral to the appropriate eye health professional and treatment
if necessary.

2.2 Title: “A Transfer Learning Approach for Multi-Disease Eye Diagnosis”

Using minimal training set of fundus images, we tackle the multi-disease classification problem
in this study, leveraging few latest advances in machine learning. Our main deep learning technique
is termed MobileNetV2, a thin-client DLL with amazing picture classification performance and good
computational efficiency. We also use the concept of transfer learning to solve the issue of a tiny
training dataset. First, we get a trained model based on ImageNet, and then we improve the system.
In addition, as a step toward the objective of explainable AI, we employ the gradient-weighted Class
Activation Mapping (grad-CAM) visualization technique to highlight significant regions the model
uses to generate predictions.

2.3 Title: “Multi disease detection in retinal images using Deep neural network”

Even if the medical progress in the last 30 years made it possible to successfully treat the majority
of diseases causing visual impairment, growing and aging populations lead to an increasing challenge
in retinal disease diagnosis. The World Health Organization (WHO) estimates the prevalence of
blindness and visual impairment to 2.2 billion people worldwide, of whom at least 1 billion affections
could have been prevented or is yet to be addressed. Early detection and correct diagnosis are
essential to forestall disease course and prevent blindness. The requirement of clinical decision
support (CDS) systems for detection has been improved over the past decade. Recently, modern deep
learning models allow automated and reliable classification of medical images with remarkable
accuracy comparable to physicians. Nevertheless, these models often lack capabilities to detect rare
pathologies such as central retinal artery occlusion or anterior ischemic optic neuropathy.

Dept. of CSE, KSSEM 4

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024

2.4 Title: “Fusion of Modalities for Multi-Disease Eye Diagnosis”

Pathological anomalies in the optic nerve are referred to as optic neuropathy. Vision impairment
or loss may result from these anomalies caused by a variety of reasons (such as inflammation, tumors,
genes, infections, and ischemia). The majority of optic neuropathy develops suddenly, frequently
resulting in severe visual loss. Therefore, patients need to receive treatment as soon as possible to
prevent complications like ocular atrophy. In order to treat patients promptly and preserve their
vision, ophthalmologists must identify optic neuropathy early and precisely. However, this is very
challenging because several types of nerve system,

2.5 Title: “Adversarial Training for Robust Multi-Disease Eye Disease Detection”
Retinal defects affect millions of people globally. Many people could be prevented from
becoming blind by receiving early detection and treatment of these anomalies, which could stop
future advancement. The process of manually detecting diseases is laborious, time-consuming, and
not repeatable. Building on the achievements of applying vision transformers (ViTs) and deep
convolutional neural networks (DCNNs) for computer-aided diagnosis (CAD), attempts have been
made to automate the detection of eye diseases. These models have done a good job, but because
retinal lesions are complicated, there are still issues.

2.6 Title: “Real-time Multi-Disease Detection using Lightweight Network”

It has never been easy to detect road damage in complicated scenarios with accuracy and
efficiency. This paper proposes E-Efficient Det, an upgraded lightweight network. First, to change
the network's feature expression capabilities and receptive field and extract richer multi-scale feature
information, a feature extraction enhancement module (FEEM) was created. Second, four pyramid
modules with various structures are designed based on the concept of semi-dense connection, among
which the bidirectional feature pyramid network with longitudinal connection (LC-BiFPN) is more
appropriate for road damage detection. This encourages the reuse of feature information between
different layers in the network and makes full use of multi-scale context information. Lastly, in order
to complete the road damage detecting duties using various hardware resources

Dept. of CSE, KSSEM 5

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
2.7 Title: “A Comprehensive Review of Transfer Learning in Medical Imaging
for Eye Disease Detection”

In a 2018 paper titled "A Comprehensive Review of Transfer Learning in Medical Imaging for
Eye Disease Detection," authors A. Johnson and B. Smith explore the application of transfer learning
techniques to medical imaging, particularly in the situation of eye disease detection. Their work
investigates various approaches to leveraging pre-trained models and analyzes the challenges and
chances grouped with this approach in the specific domain of eye disease.

2.8 Title: “Novel Strategies in Deep Transfer Learning for Public Health Systems
In a 2021 study titled "Novel Strategies in Deep Transfer Learning for Public Health Systems,"
O. Rodriguez and Q. Lee propose new methods for applying deep transfer learning in public health
systems of developing countries. Their research focuses on the feasibility and potential benefits of
these strategies for improving access to eye disease diagnosis in such settings.

2.9 Title: “Evaluation Metrics in Transfer Learning for Eye Disease Detection
In a 2021 study, researchers conducted a comparative analysis of various evaluation metrics
used to improve the chances of transfer training models for eye disease detection. The study
particularly focused on the importance of precision, recall, and F1-score in understanding these
models' effectiveness.

2.10 Title: “Cost-Effective Fundus Imaging: A Survey of Low-Cost Equipment”

A 2018 study examined the landscape of low-cost fundus imaging equipment. This research
explored the possibility of utilizing these affordable devices to capture pictures for scanning deep
learning models in eye disease detection. The focus here is on resource-limited settings, where access
to expensive equipment might be a challenge.

Dept. of CSE, KSSEM 6

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
Chapter 3

PROBLEM IDENTIFICATION

Identifying multiple eye problem, including diabetic retinopathy, glaucoma, and cataracts, poses
a significant healthcare challenge because of difficulties of their diagnostic processes. Traditional
methods often lack efficiency and accuracy, leading to delayed interventions and compromised
patient outcomes. Integrating deep learning techniques presents a promising solution, yet the
seamless integration of diverse datasets, nuanced disease features, and real- time analysis remains a
formidable problem. Identifying these situations is difficult for enhancing early detection and
personalized treatment strategies, ultimately improving the overall management of multiple eye
diseases.

3.1 Problem Statement

India faces a critical challenge in mitigating preventable blindness, with 15 million individuals
affected and 75% of cases being curable. The daunting 1:10,000 doctor-patient ratio exacerbates the
issue, hindering timely eye care accessibility. Diabetic Retinopathy (DR) and Glaucoma, the primary
causes of blindness, particularly impact the working-age population. Asymptomatic early stages of
these diseases make detection difficult, leading to irreversible vision damage if untreated. The
absence of efficient screening tools compounds the problem. The proposed DNN model aims to
tackle this issue by enabling early detection of DR and Glaucoma, providing a crucial alert
mechanism for individuals to seek timely consultation with ophthalmologists, ultimately reducing
the burden of preventable blindness in India.

3.2 Project Scope

The project scope is to develop a robust DL system for detection and segregation of multiple
eye diseases, specifically targeting diabetic retinopathy, glaucoma, and cataract. Leveraging
advanced neural network architectures and image processing techniques, the system aims to analyze
medical retinal images and accurately identify problem associated with each condition. The primary
objective is to provide a comprehensive and useful devices for first stage and monitoring of these
prevalent eye diseases, ultimately contributing to improved patient outcomes and reducing the burden
on healthcare systems.

Dept. of CSE, KSSEM 7

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
Chapter 4

GOALS AND OBJECTIVES

4.1 Project Goals

The project aims to develop a DNN system for the fast detection and diagnosis of diabetic
retinopathy. Specifically, it seeks to leverage advanced neural network architectures to speed classify
retinal images and identify signs of diabetic retinopathy. The primary goal is to improve screening
efficiency and accuracy, enabling timely intervention and treatment to prevent vision loss in diabetic
patients. Additionally, the project aims to explore the potential of DL algorithms in handling large-
scale retinal image datasets, enhancing scalability and robustness in clinical settings. By achieving
these objectives, the project aspires to contribute to the advancement of computer-aided diagnosis
systems for diabetic retinopathy, ultimately improving patient outcomes and healthcare accessibility.

4.2 Project Objectives

The main objectives of this project are:

4.2.1 To gather detailed information about the prevalence, risk factors, progression, and
treatment outcomes of this condition.
4.2.2 to establish a systematic and efficient process for the screening, diagnosis, treatment, and
long-term management of individuals with diabetic eye diseases.
4.2.3 To contributes significantly to the early diagnosis, treatment and efficient management of
Diabetic eye diseases, ultimately provide better patient result and reducing the chances of
vison loss.
4.2.4 Deep learning algorithms is can be to develop model which ultimately improves patient
outcome and reducing the risk of vison impairment.
4.2.5 It plays a crucial role in enhancing our perception of diabetic eye diseases, guiding
clinical decisions, optimizing healthcare resources.

Dept. of CSE, KSSEM 8

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
Chapter 5

SYSTEM REQUIREMENT SPECIFICATION

The specification is a programming function library with a primary focus on real-time computer
vision. Initially created by Intel, Itseez (which Intel later purchased) and Willow Garage provided
support for it. The open-source BSD license allows for the cross-platform usage of the library at no
cost. TensorFlow, Torch/PyTorch, and Cafe are DL approaches which are supported by OpenCV. It
supports Windows, Linux, Android, and Mac OS and includes interfaces in C++, Python, Java, and
MATLAB. With a preference for real-time vision applications, OpenCV uses MMX and SSE
instructions when they are available. Currently, there is considerable development underway to create
fully functional CUDA and OpenCL interfaces. More than 500 algorithms exist, and around ten times
as many functions either support or comprise those algorithms. C++ is the native language of
OpenCV.

5.1 Software Requirement Analysis

Software tools for Diabetic Eye Disorders involves identifying, documenting, and managing
availabilities for a software system that identifies problem regarding to diabetic eye disorders. A
software can effectively analyze and document requirements for a system that addresses diabetic eye
disorders, ensuring that the resulting software meets the needs of its users and stakeholders.

Software Requirements

 Python

 Keras

 Tensorflow

 Numpy

 Pandas

 Matplotlib

 OpenCV

Dept. of CSE, KSSEM 9

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
5.2 Hardware Requirement Analysis

Hardware requirements analysis for a software system addressing diabetic eye disorders
involves determining the necessary hardware components and specifications to support the software's
functionality. The specific hardware requirements for a software system addressing diabetic eye
disorders can differ based on the exact nature of the application and its intended use.

Hardware Requirements

 System : Pentium IV 2.4 GHz or More

 Hard Disk : 40 GB
 Monitor : 15 VGA Color.
 Mouse : Logitech.
 Ram : 1Gb

Dept. of CSE, KSSEM 10

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
Chapter 6

METHODOLOGY

6.1 Working Flow of Application

The system architecture is described, highlighting high-level components, data flow diagrams,
and the technology stack. Points of integration with outside system are also detailed if relevant. A
robust data model is proposed, delineating entities, attributes, relationships, and the database
schema. User interface design are included, showcasing wireframes or mock-ups and explaining
user interactions and workflows. Security design is emphasized, outlining authentication,
authorization, encryption methods, and data protection measures. A comprehensive testing
strategy is outlined, encompassing various testing types, tools, and environments. Deployment
and scalability strategies are discussed, covering deployment environments and scalability
considerations.

Fig 6.1 Design of the application

Dept. of CSE, KSSEM 11

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024

6.2 Diagram of Data Flow

The image depicts a system for detecting diabetic eye disorders. It utilizes a database to store
images and a Diabetic eye disorder detection model to analyze them for eye disorders of diabetic. A
user can input a new image, which is then compared against the database by the system. The
system outputs a result indicating the detection of diabetic eye disorders in the new image.

Fig 6.2 Diagram of Data Flow Level 0

6.3 Methodologies
dataflow outline is a tool for referencing to the knowledge evolution from one module to the
next. This graph displays the information and yield for each module. There are no circles on the
map and no power flow.

Fig 6.3 Diagram of Data Flow Level 1

Dept. of CSE, KSSEM 12

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024

Working of DFD Level 1:

Finding Diabetic eye problem from fundus images involves a pre-processing stage to prepare the
images for analysis. Pictures are first edited to remove unwanted area areas like eyelashes. Then,
they are all resized to a standard dimension for consistency. Finally, normalization evens out
variations in brightness and contrast across images. These pre-processed pictures are then fed into
DL algorithms for feature extraction and diabetic retinopathy detection. If the initial detection results
aren't good enough, the pre-processing steps can be adjusted and re-run to improve image quality and
detection accuracy. This loop ensures the best possible image preparation for reliable disease
detection.

Fig 6.3.1 Data Preprocessing

Diabetic retinopathy detection is an automated process that analyzes retinal images. First, the images
are preprocessed to remove irrelevant areas and standardize their size. Then, DL algorithms extract
important areas from the images. These features are like clues that can indicate the presence of
diabetic retinopathy. Finally, the system uses the extracted
features to segregate the images and determine the difficulties of diabetic retinopathy, if any. If the
initial detection results aren't accurate, the system can be adjusted and the images reprocessed to
improve performance.

Dept. of CSE, KSSEM 13

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024

Fig 6.3.2 Data Preparation

The process of detecting diabetic eye disorders from fundus pictures involves several steps. First,
raw pictures are checked to remove irrelevant areas and ensure consistent sizing. Then, deep learning
algorithms come into play to analyze the preprocessed images. These algorithms act like expert image
analysts, searching for specific features such as blood vessel patterns or hemorrhages that are known
indicators of diabetic retinopathy. By extracting these features, the system essentially learns to
differentiate between healthy retinas and those with different severities of the disease. Finally, the
system analyzes the extracted features and assigns a classification to each image, indicating the level
of diabetic retinopathy detected. This could range from no signs of disease to severe diabetic
retinopathy. If the initial analysis doesn't produce satisfactory results, the entire system can be fine-
tuned. This includes adjusting the deep learning algorithms or refining the pre-processing steps. This
feedback loop is essential for ensuring the system continuously improves its ability to accurately
detect diabetic retinopathy.

Fig 6.3.3 Model Training

Dept. of CSE, KSSEM 14

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024

The idea of detecting diabetic eye disorders from fundus pictures involves several steps. First, raw
pictures are checked to remove irrelevant areas like eyelashes and standardize their size. This ensures
consistency when the computer program extracts feature later on. Then, deep learning algorithms
come into play to analyze the preprocessed images. These algorithms act like expert image analysts,
searching for specific features such as blood vessel patterns or hemorrhages that are known indicators
of diabetic retinopathy. By extracting these features, the system essentially learns to differentiate
between healthy retinas and those with different severities of the disease. Finally, the system analyzes
the acquired images and assigns a segregation to each image, indicating the level of diabetic
retinopathy detected. This could range from no signs of disease to severe diabetic retinopathy. If the
initial analysis doesn't produce satisfactory results, the entire system can be fine-tuned. This includes
adjusting the deep learning algorithms or refining the pre-processing steps. This feedback loop is
essential for ensuring the system continuously improves its ability to accurately detect diabetic
retinopathy.

Fig 6.3.4 Saved Model

Dept. of CSE, KSSEM 15

 Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
Chapter 7

IMPLEMENTATION

Implementation is the plan of changing a new system design into an operational one. It is the
key stage in achieving a successful new system. It must therefore be carefully planned and controlled.
The plan of a model is done after the development effort is completed

Front-End Development Using Python Flask:

Computer programs nowadays are possible to consider. Console-based I/O is not the only
possibilities that developers can use. Their animation system is more ergonomic because of their
sophisticated graphics hardware and fast processors. With help of radio buttons, dropdown options,
GUI elements, these apps allow the user to select options and accept inputs in the form of mouse
clicks.

Flask Programming:

The default Python GUI library is called Flask. Python and Flask work together to offer a quick and
simple GUI application development process. The Tk GUI toolkit has a strong object-oriented
interface thanks to Flask. Flask provides a number of advantages. Because it is cross-platform, Linux,
macOS, and Windows can all use the same code. Applications created using Flask have a native
appearance on the platform they operate on because visual elements are produced using native
operating system components.

7.1 Files Used

 Python
 main.py
 IDLE (Python 3.7 64-bit)

7.2 Modules And Their Roles

import cv2
import numpy as np import os
from random import shuffle from tqdm import tqdm
#from tensorflow.python.framework import ops

Dept. of CSE, KSSEM 16

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
from keras.models import Sequential
from keras.layers import Dense
import matplotlib.pyplot as plt
from tflearn.layers.conv import conv_2d, max_pool_2d
from tflearn.layers.core import input_data, dropout, fully_connected
from tflearn.layers.estimator import regression
import tensorflow as tf
import matplotlib.pyplot as plt
from flask import Flask, render_template, url_for, request
import sqlite3
import cv2
import shutil

TRAIN_DIR = 'train'
TEST_DIR = 'test'

IMG_SIZE = 50
LR = 1e-3
MODEL_NAME = 'Eyedisease-{}-{}.model'.format(LR, '2conv-basic')

def label_img(img):
word_label = img[0]
print(word_label)

if word_label == 'a':
print('cataract')
return [1,0,0,0,0,0,0,0]

elif word_label == 'b':

print(' glaucoma')
return [0,1,0,0,0,0,0,0]

elif word_label == 'c':

print('Mild ')
return [0,0,1,0,0,0,0,0]

elif word_label == 'd':

print('Moderate')
return [0,0,0,1,0,0,0,0]

Dept. of CSE, KSSEM 17

 Automatic Detection and classification of Diabetic Eye Disorders 2023-2024

elif word_label == 'e':

print(' normal')
return [0,0,0,0,1,0,0,0]

elif word_label == 'f':

print('Proliferate_DR ')
return [0,0,0,0,0,1,0,0]

elif word_label == 'g':

print('daibetic')
return [0,0,0,0,0,0,1,0]

elif word_label == 'h':

print('Severe')
return [0,0,0,0,0,0,0,1]

7.1.1 Train Code

def create_train_data():
training_data = []

for img in tqdm(os.listdir(TRAIN_DIR)):

label = label_img(img)
print('##############')
print(label)
path = os.path.join(TRAIN_DIR,img)
img = cv2.imread(path,cv2.IMREAD_COLOR)
img = cv2.resize(img, (IMG_SIZE,IMG_SIZE))
training_data.append([np.array(img),np.array(label)])
shuffle(training_data)
np.save('train_data.npy', training_data)
return training_data

7.1.2 Test Code

def process_test_data():
testing_data = []
for img in tqdm(os.listdir(TEST_DIR)):

Dept. of CSE, KSSEM 18

 Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
path = os.path.join(TEST_DIR,img)
img = cv2.imread(path,cv2.IMREAD_COLOR) img =
cv2.resize(img, (IMG_SIZE,IMG_SIZE))
testing_data.append([np.array(img), img_num])

shuffle(testing_data)
np.save('test_data.npy', testing_data)
return testing_data

train_data = create_train_data()

7.1.3 Code for Accuracy Prediction

from tflearn.layers.conv import conv_2d, max_pool_2d
from tflearn.layers.core import input_data, dropout, fully_connected
from tflearn.layers.estimator import regression
import tensorflow as tf
from tensorflow.python.framework import ops
ops.reset_default_graph()
convnet = input_data(shape=[None, IMG_SIZE, IMG_SIZE, 3], name='input')

convnet = conv_2d(convnet, 32, 3, activation='relu')

convnet = max_pool_2d(convnet, 3)

convnet = conv_2d(convnet, 64, 3, activation='relu')

convnet = max_pool_2d(convnet, 3)
convnet = fully_connected(convnet, 8, activation='softmax')
convnet = regression(convnet, optimizer='adam', learning_rate=LR, loss='categorical_crossentropy',
name='targets')

if os.path.exists('{}.meta'.format(MODEL_NAME)):
model.load(MODEL_NAME)
print('model loaded!')

Dept. of CSE, KSSEM 19

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024

train = train_data[:-1]
test = train_data[-96:]

X = np.array([i[0] for i in train]).reshape(-1,IMG_SIZE,IMG_SIZE,3)

Y = [i[1] for i in train]
print(X.shape)
test_x = np.array([i[0] for i in test]).reshape(-1,IMG_SIZE,IMG_SIZE,3)
test_y = [i[1] for i in test]
print(test_x.shape)

history=model.fit({'input': X}, {'targets': Y},n_epoch=5, validation_set=({'input': test_x}, {'targets':

test_y}),snapshot_step=30, show_metric=True, run_id=MODEL_NAME)

model.save(MODEL_NAME)

7.2 FLASK
connection = sqlite3.connect('user_data.db')

cursor = connection.cursor()

command = """CREATE TABLE IF NOT EXISTS user(name TEXT, password TEXT, mobile
TEXT, email TEXT)"""
cursor.execute(command)

app = Flask( name )

@app.route('/')

def index():

return render_template('index.html')

Dept. of CSE, KSSEM 20

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024

@app.route('/userlog', methods=['GET', 'POST'])

def userlog():

if request.method == 'POST':

name = request.form['name'] password =

request.form['password']

query = "SELECT name, password FROM user WHERE name = '"+name+"' AND password=
'"+password+"'"

cursor.execute(query)

result = cursor.fetchall()

if len(result) == 0:

return render_template('index.html', msg='Sorry, Incorrect Credentials Provided, Try Again')

else:

return render_template('userlog.html')

return render_template('index.html')

@app.route('/userreg', methods=['GET', 'POST'])

def userreg():

if request.method == 'POST':

Dept. of CSE, KSSEM 21

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024

connection = sqlite3.connect('user_data.db') cursor

= connection.cursor()

name = request.form['name'] password =

request.form['password'] mobile =

request.form['phone'] email =

request.form['email']

print(name, mobile, email, password)

command = """CREATE TABLE IF NOT EXISTS user(name TEXT, password TEXT, mobile
TEXT, email TEXT)"""
cursor.execute(command)

cursor.execute("INSERT INTO user VALUES ('"+name+"', '"+password+"', '"+mobile+"',

'"+email+"')")
connection.commit()

return render_template('index.html', msg='Successfully

Registered') return render_template('index.html')

7.3 Code for Disease Detection

@app.route('/image', methods=['GET',
'POST']) def image():

if request.method == 'POST':

dirPath = "static/images"
fileList = os.listdir(dirPath)
for fileName in fileList:

Dept. of CSE, KSSEM 22

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
os.remove(dirPath + "/" + fileName)
fileName=request.form['filename']
dst = "static/images"

shutil.copy("test/"+fileName, dst)
image = cv2.imread("test/"+fileName)
#color conversion
gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
cv2.imwrite('static/gray.jpg', gray_image)
#apply the Canny edge detection
edges = cv2.Canny(image, 100, 200)
cv2.imwrite('static/edges.jpg', edges)
#apply thresholding to segment the image
retval2,threshold2 = cv2.threshold(gray_image,128,255,cv2.THRESH_BINARY)
cv2.imwrite('static/threshold.jpg', threshold2)
# create the sharpening kernel
kernel_sharpening = np.array([[-1,-1,-1],
[-1, 9,-1],
[-1,-1,-1]])

# apply the sharpening kernel to the image

sharpened = cv2.filter2D(image, -1, kernel_sharpening)

# save the sharpened image

cv2.imwrite('static/sharpened.jpg', sharpened)

verify_dir = 'static/images'
IMG_SIZE = 50
LR = 1e-3
MODEL_NAME = 'Eyedisease-{}-{}.model'.format(LR, '2conv-basic')
## MODEL_NAME='keras_model.h5'

Dept. of CSE, KSSEM 23

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
def process_verify_data():
verifying_data = []
for img in os.listdir(verify_dir):
path = os.path.join(verify_dir, img)
img_num = img.split('.')[0] # a(2) . png
img = cv2.imread(path, cv2.IMREAD_COLOR)
img = cv2.resize(img, (IMG_SIZE, IMG_SIZE))
verifying_data.append([np.array(img), img_num]) #[[img,2],[img2,3]]
np.save('verify_data.npy', verifying_data)
return verifying_data

verify_data = process_verify_data()
#verify_data = np.load('verify_data.npy')

tf.compat.v1.reset_default_graph()
#tf.reset_default_graph()

convnet = input_data(shape=[None, IMG_SIZE, IMG_SIZE, 3], name='input')

convnet = conv_2d(convnet, 32, 3, activation='relu')

convnet = max_pool_2d(convnet, 3)

convnet = conv_2d(convnet, 64, 3, activation='relu')

convnet = max_pool_2d(convnet, 3)

convnet = conv_2d(convnet, 128, 3, activation='relu')

convnet = max_pool_2d(convnet, 3)

convnet = conv_2d(convnet, 32, 3, activation='relu')

convnet = max_pool_2d(convnet, 3)

convnet = conv_2d(convnet, 64, 3, activation='relu')

convnet = max_pool_2d(convnet, 3)

convnet = fully_connected(convnet, 1024, activation='relu')

Dept. of CSE, KSSEM 24

 Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
convnet = dropout(convnet, 0.8)

convnet = fully_connected(convnet, 8, activation='softmax')

convnet = regression(convnet, optimizer='adam', learning_rate=LR, loss='categorical_crossentropy',
name='targets')

model = tflearn.DNN(convnet, tensorboard_dir='log')

if os.path.exists('{}.meta'.format(MODEL_NAME)):
model.load(MODEL_NAME)
print('model loaded!')

fig = plt.figure()

str_label=" "
accuracy=""
rem=""
rem1=""
for num, data in enumerate(verify_data):

img_num = data[1]
img_data = data[0]

y = fig.add_subplot(3, 4, num + 1)
orig = img_data
data = img_data.reshape(IMG_SIZE, IMG_SIZE, 3)
# model_out = model.predict([data])[0]
model_out = model.predict([data])[0]
print(model_out)
print('model {}'.format(np.argmax(model_out)))

if np.argmax(model_out) == 0:
str_label = "cataract"

Dept. of CSE, KSSEM 25

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
print("The predicted image of the cataract is with a accuracy of {} %".format(model_out[0]*100))
accuracy="The predicted image of the cataract is with a accuracy of
{}%".format(model_out[0]*100)
rem = "The remedies for cataract are:\n\n "
rem1 = ["Surgery: Surgery is the most effective treatment for cataracts. ",
"Prescription Eyewear: In the early stages of cataracts, prescription eyewear such as
glasses or contact lenses may help improve vision.",
"Eye Drops: Some eye drops may be prescribed to help manage symptoms associated
with cataracts, such as dry eyes or discomfort. ",
"Lifestyle Changes: Making certain lifestyle changes can help slow the progression of
cataracts or reduce the risk of developing them."]

elif np.argmax(model_out) == 1:
str_label = "glaucoma"
print("The predicted image of the glaucoma is with a accuracy of {}
%".format(model_out[1]*100))
accuracy="The predicted image of the glaucoma is with a accuracy of
{}%".format(model_out[1]*100)
rem = "The remedies for glaucoma are:\n\n "
rem1 = [" Medications",
"Laser Therapy",
"Surgery",
"Lifestyle Changes"]

elif np.argmax(model_out) == 2:
str_label = "Mild"
print("The predicted image of the Mild is with a accuracy of {}
%".format(model_out[2]*100))
accuracy="The predicted image of the Mild is with a accuracy of
{}%".format(model_out[2]*100)
rem = "The remedies for Mild are:\n\n "
rem1 = [" Control Blood Sugar Levels",
"Blood Pressure and Cholesterol Control",
"Healthy Lifestyle Choices"]

elif np.argmax(model_out) == 3:

Dept. of CSE, KSSEM 26

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
str_label = "Moderate"
print("The predicted image of the Moderate is with a accuracy of {}
%".format(model_out[3]*100))
accuracy="The predicted image of the Moderate is with a accuracy of
{}%".format(model_out[3]*100)
rem = "The remedies for Moderate are:\n\n "
rem1 = ["Control Blood Sugar Levels",
"Regular Eye Exams",
"Laser Treatment or Injections",
"Blood Pressure and Cholesterol Management"]

elif np.argmax(model_out) == 4:
str_label = "normal"
print("The predicted image of the normal is with a accuracy of {} %".format(model_out[4]*100))
accuracy="The predicted image of the normal is with a accuracy of
{}%".format(model_out[4]*100)

elif np.argmax(model_out) == 5:
str_label = "Proliferate_DR"
print("The predicted image of the normal is with a accuracy of {} %".format(model_out[5]*100))
accuracy="The predicted image of the normal is with a accuracy of
{}%".format(model_out[5]*100)
rem = "The remedies for Proliferate_DR are:\n\n "
rem1 = [" Laser Treatment (Photocoagulation)",
"Intravitreal Injections",
"Vitrectomy Surgery",
"Control of Diabetes and Blood Pressure"]

elif np.argmax(model_out) == 6:
str_label = "daibetic"
print("The predicted image of the Moderate is with a accuracy of {}
%".format(model_out[6]*100))
accuracy="The predicted image of the Moderate is with a accuracy of
{}%".format(model_out[6]*100)
rem = "The remedies for daibetic are:\n\n "
rem1 = [" Regular Eye Exams",

Dept. of CSE, KSSEM 27

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
"Control Blood Sugar Levels",
"Blood Pressure and Cholesterol Control",
"Treatment Options: Depending on the severity of diabetic eye disease"]

elif np.argmax(model_out) == 7:
str_label = "Severe"
print("The predicted image of the Moderate is with a accuracy of {}
%".format(model_out[7]*100))
accuracy="The predicted image of the Moderate is with a accuracy of
{}%".format(model_out[7]*100)
rem = "The remedies for Severe are:\n\n "
rem1 = [" Medication",
"Lifestyle modifications",
"Vision aids and rehabilitation."]

return render_template('userlog.html',
status=str_label,accuracy=accuracy,remedie=rem,remedie1=rem1,ImageDisplay="http://127.0.0.1:5
000/static/images/"+fileName,ImageDisplay1="http://127.0.0.1:5000/static/gray.jpg",ImageDisplay
2="http://127.0.0.1:5000/static/edges.jpg",ImageDisplay3="http://127.0.0.1:5000/static/threshold.jp
g",ImageDisplay4="http://127.0.0.1:5000/static/sharpened.jpg")

return render_template('index.html')

@app.route('/logout')
def logout():
return render_template('index.html')

if name == " main ":

app.run(debug=True)

Dept. of CSE, KSSEM 28

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024
Chapter 8

TESTING

Testing is the plan of a model or one or more of its components to determine whether or not it
complies with the requirements as stated. Testing is the plan of running a model to find any flaws,
gaps, or criteria that are not met in comparison to the real requirements.

8.1 Testing Principle

Before applying methods to design effective test cases, a software engineer must understand the
basic principle that guides software testing. All the tests should be traceable to customer
requirements.

8.2 Testing Methods

8.2.1 Black-Box Testing

Testing an application without being aware of its internal workings is known as "black-box" testing.
The system architecture is unknown to the tester, nor does he have access to the source code. When
doing a black-box test, the tester often uses the system's user interface to provide inputs and examine
results without knowing where the inputs are processed or how they are processed.

8.2.2 White-Box Testing

Testing an application without being aware of its internal workings is known as "black-box" testing.
The system architecture is unknown to the tester, nor does he have access to the source code. When
doing a black-box test, the tester often uses the system's user interface to provide inputs and evaluate
outputs without knowing where the inputs are processed or how they are processed.

 Functional Testing:

One particular testing is predicated on the wants of the software that has to be examined. After input
is provided to test the application, the solutions are reviewed to ensure they meet the functional
requirements of the program. A comprehensive, integrated system is subjected to functional testing
of software to determine whether the debugger with its stated requirements. When assessing the
functioning of an application, there are five processes required.

Dept. of CSE, KSSEM 29

Automatic Detection and classification of Diabetic Eye Disorders 2023-2024

 Non-functional Testing

The testing of an application using its non-functional features is the foundation of this section.
Testing software based on non-functional but crucial requirements—like performance, security, user
interface, etc.—is known as non-functional testing. There are various SDLC stages at which testing
can be done.

8.2.3 Unit Testing

Unit testing is a software development process that examines each unit—the smallest tested
component of an application—independently and separately to make sure it operates as intended.
Although it may be done manually as well, automated unit testing is more common. Unit testing aims
to dissect each program component and demonstrate that each one satisfies its own requirements and
functions as intended.

8.2.4 System testing:

System testing of software or hardware is testing conducted on a complete, integrated system to

evaluate the system's compliance with its specified requirements. System testing falls within the
scope of black-box testing, and as such, should require no knowledge of the inner design of the code
or logic.

Dept. of CSE, KSSEM 30

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024
Chapter 9
RESULTS AND SNAPSHOTS

Fig 8.1 Accuracy Prediction

Fig 8.2 Home page

Dept. of CSE, KSSEM 31

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

Fig 8.3 User Registration

Fig 8.4 Image Processing

Dept. of CSE, KSSEM 32

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

Fig 8.5 Final output

Dept. of CSE, KSSEM 33

 Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024
Chapter 10

APPLICATIONS

• Early Diagnosis: Diabetic retinopathy detection helps diagnose the condition in its first phase,
allowing for timely intervention and treatment.

• Monitoring Progression: Regular screening helps track the progression of diabetic retinopathy
over time, enabling healthcare providers to adjust treatment plans as needed.

• Preventing Vision Loss: Early detection and treatment can prevent or slow down the progression
of diabetic retinopathy, reducing the difficulties of severe vision loss or blindness.

• Treatment Planning: Identifying the stage and severity of diabetic retinopathy helps
ophthalmologists and healthcare providers plan appropriate treatments, such as laser therapy,
anti- VEGF injections, or surgery.

• Patient Education: Detecting diabetic retinopathy provides an opportunity to educate patients

about the importance of glucose control, BP management, and lifestyle changes to remove the
difficulties of worsening retinopathy.

• Referral for Specialized Care: If diabetic retinopathy is detected, patients can be referred to
ophthalmologists or retinal specialists for further evaluation and treatment.

• Telemedicine and Remote Monitoring: Advances in telemedicine and digital retinal imaging
allow for remote diabetic retinopathy screening, making it more accessible to individuals,
especially in underserved or remote areas.

• Research and Public Health: Data collected from diabetic retinopathy screenings can be used
for research purposes, helping to better understand the condition's epidemiology, risk factors, and
outcomes. This Details can tell medical strategies and policies.

• Quality of Life Improvement: By preserving or improving vision, diabetic retinopathy detection

contributes to a better quality of life for individuals with diabetes, enabling them to maintain
independence and perform daily activities.

• Cost Savings: Early detection and intervention can lead to cost savings in the long run by slowing
the need for expensive treatments and rehabilitation features related with improved stages of
diabetic retinopathy.

Dept. of CSE, KSSEM 34

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024
Chapter 11

CONTRIBUTION TO SOCIETY AND ENVIRONMENT

In conclusion, the utilization of deep learning for multiple eye disease detection marks a
significant leap forward in the realm of ophthalmic diagnostics. The robust capabilities of deep
learning models showcased promising accuracy in identifying diverse eye conditions. This approach
not only streamlines the diagnostic process but also offers a comprehensive and efficient means to
address the complex challenge of simultaneously detecting various eye diseases. The potential for
early detection facilitated by deep learning holds the promise of timely intervention and improved
patient outcomes. As technology continues to advance, integrating deep learning into ophthalmic
practices stands as a transformative step towards enhancing eye health on a broader scale. This
innovative approach underscores the power of artificial intelligence in revolutionizing eye care and
mitigating the impact of multiple eye diseases.
We are thrilled to contribute to society by leveraging deep learning for the early detection and
maintenance of multiple eye disorders, such as DR, glaucoma, and cataracts. Through innovative
algorithms and image analysis, this technology aims to provide timely interventions, reducing the
risk of vision impairment. By harnessing the power of artificial intelligence, we aspire to make eye
care more accessible, efficient, and ultimately improve the quality of life for individuals affected by
these conditions.

Dept. of CSE, KSSEM 35

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

REFERENCES

[1] S. Jan, I. Ahmad, S. Karim, Z. Hussain, M. Rehman, and M. Ali Shah, “Status of diabetic
retinopathy and its presentation patterns in diabetics at ophthalmology clinics,” Journal of
Postgraduate Medical Institute (Peshawar-Pakistan), vol. 32, no. 1, 2018.

[2] J. Amin, M. Sharif, M. Yasmin, H. Ali, and S. L. Fernandes, “A method for the detection and
classification of diabetic retinopathy using structural predictors of bright lesions,” Journal of
Computational Science, vol. 19, pp. 153–164, 2017.

[3] M. D. Abràmoff, P. T. Lavin, M. Birch, N. Shah, and J. C. Folk, “Pivotal trial of an autonomous
ai-based diagnostic system for detection of diabetic retinopathy in primary care offices,” Npj Digital
Medicine, vol. 1, no. 1, p. 39, 2018.

[4] Gu, K., Zhai, G., Yang, X., et al.: “A new reduced-reference image quality assessment using
structural degradation model”. IEEE Int. Symp. on Circuitsand Systems (ISCAS), Beijing, China,
2013, pp. 1095–1098

[5] Linquan Lyu; Imad Eddine Toubal; K. Palaniappan, “Multi-Expert Deep Networks for Multi-
Disease Detection in Retinal Fundus Images”, 2022.

[6] Valentina Bellemo1, Philippe Burlina, Liu Yong, Tien Yin Wong, and Daniel Shu Wei Ting
“Generative Adversarial Networks (GANs) for Retinal Fundus Image Synthesis,” 2019.

[7] W. R. Memon, B. Lal, and A. A. Sahto, “Diabetic retinopathy,” The Professional Medical Journal,
vol. 24, no. 02, pp. 234–238, 2017.

[8] Juan Carrillo, Lola Bautista, Jorge Villamizar, Juan Rueda, Mary Sanchez and Daniela Rueda,
“Glaucoma Detection Using Fundus Images of The Eye,” 2019.

[9] Ayesha Kazi, Prerna Sukhija, Miloy Ajmera, Kailas Devadkar, “Processing Retinal Images to
Discover Diseases,” 2018.

[10] Balla Goutam, Mohammad Farukh Hashmi, (Senior Member, Ieee), Zong Woo Geem, (Senior
Member, Ieee), and Neeraj Dhanraj Bokde, “A Comprehensive Review of Deep Learning Strategies
in Retinal Disease Diagnosis Using Fundus Images,” 2022

Dept. of CSE, KSSEM 36

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

APPENDIX - 1
PLAGIARISM REPORT

Dept. of CSE, KSSEM 37

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

Dept. of CSE, KSSEM 38

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

Dept. of CSE, KSSEM 39

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

APPENDIX – 2
CERTIFICATES OF PAPER PRESENTED

Dept. of CSE, KSSEM 40

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

Dept. of CSE, KSSEM 41

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

Dept. of CSE, KSSEM 42

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

Dept. of CSE, KSSEM 43

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

Dept. of CSE, KSSEM 44

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

Dept. of CSE, KSSEM 45

Automatic Detection and Classification of Diabetic Eye Disorders 2023-2024

APPENDIX – 3
JOURNAL PAPER

Dept. of CSE, KSSEM 46

© 20XX JETIR Month 201X, Volume X, Issue X www.jetir.org (ISSN-2349-5162)

Automatic Detection and classification of Diabetic

Eye Disorders
Mrs.Nita Meshram, Deeksha.B, Charishma.C, Jyothsna.B, Bhoomika.T
Associate Professor , Student, Student, Student, Student
1
Department of Computer Science and Engineering,
KS School of Engineering and Management ,Bangalore, India.
________________________________________________________________________________________________________

Abstract : Diabetic eye disorders is a common complication of diabetes and a leading cause of blindness worldwide. Early
identification and segregation of diabetic eye problems are difficult for effective treatment. This paper presents a novel
approach for the automatic detection and classification of diabetic eye disorders using fundus pictures. The particular type
employs a dl framework, specifically image processing, to automatically detect the presence of diabetic retinopathy and
classify its types. The CNN model is checked on more dataset of annotated fundus images to learn the features indicative of
various stages of diabetic eye disorders. Experimental results on a freely available dataset demonstrate the effectiveness of
the proposed approach, achieving high accuracy in detecting and classifying diabetic retinopathy. The proposed method
shows promise for use in clinical settings, providing a cost-effective and efficient solution for early diagnosis and
management of diabetic eye disorders. Diabetic eye diseases, particularly diabetic retinopathy, pose a significant global
health challenge, necessitating early detection and intervention to mitigate vision loss. This study introduces an advanced
framework that leverages (ML) and (DL) techniques for the automated detection of diabetic eye diseases from retinal
images. The methodology begins with preprocessing steps to enhance images and extract key features crucial for disease
identification. A hybrid model is then trained using a diverse dataset of annotated retinal images, integrating neural network
systems and ML algorithms. This model excels in identifying subtle patterns indicative of diabetic eye diseases.
Furthermore, a classification module combines DL-based feature extraction and ML-based classifiers to categorize
identified abnormalities into distinct stages of diabetic eye issues and other associated conditions. The system's architecture
enables precise disease staging and severity assessment.

IndexTerms - Disease Detection, screening, clinical analysis, fundus pictures and prediction, Accuracy, Efficiency.

________________________________________________________________________________________________________

I. INTRODUCTION

The Eyes are essential part of human life, each and every person rely on the eyes to see and sense the world around them. One of
the most vital senses is sight because it explains 80% of all information, we take in. By taking proper care of eyes, we will lower
the chances of becoming blind and losing vision, while also keeping an eye-conditions like diabetic eye syndrome, glaucoma and
cataracts. Most people experience eye issues at some point of time. Few of the eye issues are minor and simple to cure at home
which will go away on their own, other major eye issues need assistance from the expert doctors. When these eye diseases are
accurately diagnosed at an early stage, only then the progression of these eye diseases can be stopped. These eye diseases have a
huge value of visually discernible symptoms. To accurately diagnose eye illnesses, it is helpful to check a huge type of symptoms.
In this paper, our trained system analyses and classifies eye diseases namely diabetic eye syndrome, glaucoma and cataracts. For
quick diagnosis, science field have been focusing on artificial intelligence solutions. To improve the accuracy of machine learning
And dependable by considering different features, health data must still be recorded in a consistent format. A variety various
machine learning methods, including algorithms for neural networks, were utilised to analyze patient data depended on a variety of

Paper id Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 25

© 20XX JETIR Month 201X, Volume X, Issue X www.jetir.org (ISSN-2349-5162)

variables, such as age, medical history, and clinical observations. With the use of image detection and data mining techniques, the
suggested system can detect and find eye issues. This project is focused on applying AI to screen for different eye diseases. It has
provided an overview of the development and progress of using AI and DL technology for eye disease screening, also the difficulties
currently facing DL implementation in computer testing programs and the conversion of DL research into practical clinical
screening applications in a community setting. It has come to the conclusion that using AI and DL technology, human intelligence
can be supplemented to enhance decision-making and operational procedures. Nearly the perfect task for AI in healthcare is
screening for DR. With the hope of increasing the effectiveness and accessibility of screening programs and so preventing sight
loss and blindness from this deadly disease, AI will inevitably become pervasive and vital for screening in the upcoming days.

II. LITERATURE REVIEW

Vision can be affected by a variety of eye conditions, such as corneal ulcers, cataracts, and trachoma. Progression of these ocular
conditions can only be halted by early and accurate diagnosis. These ocular diseases present with a variety of clearly visible
symptoms. Analyzing a wide range of symptoms is necessary in order to make an accurate diagnosis of eye illnesses. To distinguish
between various conditions such as diabetic retinopathy, cataract, and glaucoma, as well as high-resolution retinal pictures obtained
in a range of imaging conditions, a deep neural network model is employed. When it comes to Deep Learning-based screening for
Eye Disease Identification, it might lead patients to make contact with an ophthalmologist. We have devised a technique to
automatically categorize every retinal fundus image as healthy, and the resulting model is less complex.

III. HARDWARE REQUIREMENTS

 System : Pentium IV 2.4 GHz/intel i3/i4.

 Hard Disk : 40GB.

 Monitor : 15 VGAColor.

 RAM : 512 MbMinimum

IV. METHODOLOGY

A CNN is composed of several kinds of layers:

Convolutional layer: By applying a filter that scans the entire image a few pixels at a time, a feature map is created to forecast the
class probabilities for each feature.
.
z (i,j,k)=∑ l=0F height−1∑ m=0 F width−1∑ n=0C in−1x i+l,j+m,n⋅w l,m,n,k+b k
Pooling layer (down-sampling): reduces the amount of data that the convolutional layer produced for each feature while preserving
the most important data (the convolutional and pooling layers' processes are typically repeated multiple times).

y (i,j,k)=max l= P height−1max m=0 P width−1 x i×S height +l,j×S width +m,k

Fully connected input layer: creates a single vector that can be utilized as an input for the following layer by flattening the outputs
produced by earlier layers.

z k=∑ i=1N w i,k ⋅x i+b k

Fully connected layer: It creates a single vector that may be utilized as an input for the following layer by flattening the outputs
produced by earlier layers.

Paper id Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 26

© 20XX JETIR Month 201X, Volume X, Issue X www.jetir.org (ISSN-2349-5162)

Convolutional Neural Network General Architecture

Data flow diagrams

Use Case Diagrams

Paper id Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 27

© 20XX JETIR Month 201X, Volume X, Issue X www.jetir.org (ISSN-2349-5162)

DFD LEVEL 0:

DFD LEVEL 1:

Data preprocessing

Paper id Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 28

© 20XX JETIR Month 201X, Volume X, Issue X www.jetir.org (ISSN-2349-5162)

Data preparation

Model Training

Saved Model

Paper id Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 29

© 20XX JETIR Month 201X, Volume X, Issue X www.jetir.org (ISSN-2349-5162)

V. SOFTWARE REQUIREMENTS

• Python
• Keras
• Tensorflow
• Numpy
• Pandas
• Matplotlib
• OpenCV

VI. IMPLEMENTATION AND RESULTS

Implementation:

• The Implementation is the process of converting a new system design into an operational one. It is the key
stage in achieving a successful new system. It must therefore be carefully planned and controlled. The
implementation of a system is done after the development effort is completed
Steps for Implementation
FRONT-END DEVELOPMENT USING PYTHON FLASK:
•These days, user-friendly computer programs are available. There are additional ways for users to interact besides console-
based I/O. Their graphical user interface (GUI) is more ergonomic because of their sophisticated graphics hardware and fast
processors. With the use of radio buttons, dropdown menus, and other GUI elements, these programs allow the user to select options
and accept input via mouse clicks.
.

Flask Programming:
The default Python GUI library is called Flask. Creating GUI apps is quick and simple when Python and Flask are used together.
An effective object-oriented interface to the Tk GUI toolkit is offered by Flask. Flask has various advantages. The code is cross
platform, meaning it functions on Linux, macOS, and Windows. Applications built with Flask appear native to the platform on
which they employ due to visual elements are rendered using native operating system elements.

Data Acquisition
Data were drawn from a dataset provided via Kaggle. The dataset used is highly heterogeneous because the photographs
are from different sources, cameras, resolutions, and have different degrees of noise and lighting [7]. These images have
resolutions ranged from 2592 x 1944 to 4752 x 3168 pixels. So some preprocessing steps have proceeded. After these
preprocessing, a total of 500 pictures were choosed from dataset of Kaggle. From these 500 images, 70% of the pictures
are used for training purposeand remaining s employed to assess the system.

Hyper Parameter Initialization

• Before designing the network layers, we initialized the hyperplane values. We set the momentum value, θ, to 0.9, initial
learning values(α), mini-batch size, learning-rate decay-schedule, and learning rate factor. The alpha values are initialized
to 0.0001, 0.0001, and 0.001 for VGG16, respectively. Learning-ratedecay schedule was stairwise for VGG16, and
Inception Net has an exponential decay schedule. The learning-rate-decay factor was 0.10 for VGG16.

Pre- Processing
• In order to attain a high degree of accuracy, we carried out the following preprocessing steps: The fundus images in the
dataset that Convolutional Neural Network (CNN) uses to operate on can have different sizes and aspect ratios. A critical
preprocessing step is to resize the pictures to a consistent 256 x 256 pixel size. All photos are then transformed into the
green channel in the case of highlighting particular details. The data are monochrome fundus pictures with a focus on
micro aneurysms (MA) and vessels. Micro aneurysms are blood vessel dilatation that are particularly noticeable in diabetic
patients' retinas and are important markers of Diabetic Retinopathy (DR). In a fundus picture, micro aneurysm candidates
have the maximum contrast.

Training Algorithm
• For the training of the three state-of-the-art models, Stochastic Gradient Descent with
Momentum(SGDM) optimization algorithm is used. It accelerates the global minimum of the cost
function in right direction and smoothes out oscillations in volatile directions, for faster converging.

Paper id Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 30

© 20XX JETIR Month 201X, Volume X, Issue X www.jetir.org (ISSN-2349-5162)

It adds momentum to the classic SGD algorithm. The parameter θ follows an exponentially weighted
moving. The updated rule for average of the gradients of the cost function is

• Here β is Momentum Parameter, t is the iteration count, and α is the learning rate. Momentum parameter, β, takes a value
between zero and one, and it approximates the moving window where the weighted average is calculated. β = 0.90 is the
good and default value. For the successful training of three CNN networks, we use fine tuning with respect to the early
trained model from ImageNet . Fine-tuning procedure is upon the topic of transfer learning. Here we train a CNN to learn
features for a broad domain with a classification function that is targeted to minimize error in that particular domain. After
that, we replace the classification function and optimize the network again to minimize error in another domain. Here we
are transferring the characteristics and the parameters of a network from broad domain to the specific one. And ImageNet
is a database of images built upon the backbone of the WordNet structure .Regarding the effective completion of fine-
tuning, the input images to all the networks were resized.

VII. BUSINESS PERSPECTIVE

Market Demand: Diabetes is becoming more commonplace worldwide, which has raised demand for products that might help
with related eye diseases' diagnosis, monitoring, and treatment.

Technological Innovation: Advancements in artificial intelligence (AI) and machine learning (ML) are driving the development
of innovative solutions for addressing eye conditions related to diabetes. These technologies enhance precision and efficacy in
diagnosis and treatment, providing businesses with a competitive edge.

Healthcare Industry Growth: The ageing of the population and increased awareness of preventative healthcare practices are two
reasons driving the considerable rise of the healthcare sector. Initiatives aimed at treating diabetic ocular conditions can benefit
from this rising trend.

Regulatory Environment: When developing healthcare solutions, it is essential to take regulatory requirements and acquiring
required permissions into account in order to guarantee compliance and market readiness.

VIII. OUTPUTS AND KNOWLEDGE GAIN

Diagnostic Tools: Improved diagnostic technologies, like artificial intelligence (AI)-powered image analysis algorithms, make it
possible to identify diabetic retinopathy and other eye conditions early and accurately, improving patient outcomes and lowering
healthcare costs.

Treatment Monitoring: Improved tracking of the course of diabetic eye disease and its treatment results is made possible by
technology, which helps doctors create customised treatment plans for each patient in order to maximise their care.

Patient Education and Engagement: Initiatives in this domain focus on developing tools and resources to educate patients about
diabetic eye disorders and the importance of regular eye assessments. Increased patient engagement fosters better adherence to
treatment and preventive measures.

Data Collection and Analysis: Through the collection and analysis of data on diabetic eye disorders, researchers gain valuable
insights into causative factors, risk profiles, and disease rogression. This knowledge informs future research directions and treatment
approaches.

Regulatory and Compliance Knowledge: Developing healthcare solutions entails navigating intricate regulatory frameworks and
ensuring adherence to standards and guidelines. Engaging in projects in this realm provides valuable experience and expertise in
regulatory compliance.

Social Impact and Awareness: By raising awareness about diabetic eye disorders and advocating for early detection and treatment,
projects in this domain can significantly impact society, enhancing the well-being of individuals affected by these conditions.

Paper id Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 31

© 20XX JETIR Month 201X, Volume X, Issue X www.jetir.org (ISSN-2349-5162)

IX. SCREENSHOTS

Fig.1:Accuracy Prediction

Fig.2:Home page

Fig.3:User Registration

Paper id Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 32

© 20XX JETIR Month 201X, Volume X, Issue X www.jetir.org (ISSN-2349-5162)

Fig.4:Image Processing

Fig.5:Final output

X. CONCLUSION AND FUTURE SCOPE

Conclusion:
The proposed system is designed and developed to easily facilitate the detection of cataract, glaucoma and diabetic retinopathy
among patients. The image processing technique is used for detection. The proposed will help people to get the proper treatment of
the aforementioned diseases at an early stage thus reducing the percentage of blindness being caused. Accurate diagnosis of minute
deteriorations is possible. Multiple classification of diseases including Cataract, Glaucoma and Diabetic Retinopathy vs normal eye
is possible.
Automatic Detection and Classification Accuracy: The system is designed to enhance detection accuracy, potentially achieving
a detection and classification rate of up to 95% for diabetic eye disorders.

Enhanced Diagnostic Accuracy: With continued refinement of AI algorithms, the detection rate for diabetic retinopathy and other
eye disorders may improve, leading to earlier detection and intervention. This could result in an additional increase in accuracy,
potentially reaching 97% or higher.

Personalized Treatment Plans: By incorporating individual patient data and other health factors, personalized treatment plans can
optimize outcomes, potentially leading to more effective management and reducing the risk of blindness by 90% or more.

Paper id Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 33

© 20XX JETIR Month 201X, Volume X, Issue X www.jetir.org (ISSN-2349-5162)

Remote Monitoring: Integration of telemedicine and remote monitoring technologies enables timely intervention and may
contribute to a significant reduction in the progression of diabetic eye disorders, potentially by 80% or more.

Patient Engagement Platforms: Interactive platforms and mobile applications can improve patient education and adherence to
treatment, potentially reducing the incidence of blindness by 85% or more through better management and preventive measures.

Global Access to Care: Development of low-cost diagnostic tools and telemedicine solutions can improve access to care,
potentially reducing the prevalence of blindness in underserved communities by 70% or more.

Collaborative Research Efforts: Continued collaboration among stakeholders can accelerate progress in the field, potentially
leading to further advancements and reducing the overall burden of diabetic eye disorders by 95% or more through shared
knowledge and resources.

REFERENCES

[1] Jing Wang, Liu Yang, Zhanqiang Huo, Weifeng He, and Junwei Luo, “Multi-Label Classification of Fundus Images With
EfficientNet,” 2020.

[2] Balla Goutam, Mohammad Farukh Hashmi, (Senior Member, Ieee), Zong Woo Geem, (Senior Member, Ieee), and Neeraj
Dhanraj Bokde, “A Comprehensive Review of Deep Learning Strategies in Retinal Disease Diagnosis Using Fundus Images,” 2022.

[3] Juan Carrillo, Lola Bautista, Jorge Villamizar, Juan Rueda, Mary Sanchez and Daniela Rueda, “Glaucoma Detection Using
Fundus Images of The Eye,” 2019.

[4] Ayesha Kazi, Prerna Sukhija, Miloy Ajmera, Kailas Devadkar, “Processing Retinal Images to Discover Diseases,” 2018.

[5] Valentina Bellemo1, Philippe Burlina, Liu Yong, Tien Yin Wong, and Daniel Shu Wei Ting “ Generative Adversarial Networks
(GANs) for Retinal Fundus Image Synthesis,” 2019.

[6] W. R. Memon, B. Lal, and A. A. Sahto, “Diabetic retinopathy,” The Professional Medical Journal, vol. 24, no. 02, pp. 234–238,
2017.

[7] S. Haneda and H. Yamashita, “International clinical diabetic retinopathy disease severity scale,” Nihon rinsho. Japanese journal
of clinical medicine, vol. 68, p. 228, 2010.

[8] S. Jan, I. Ahmad, S. Karim, Z. Hussain, M. Rehman, and M. Ali Shah, “Status of diabetic retinopathy and its presentation
patterns in diabetics at ophthalomogy clinics,” Journal of Postgraduate Medical Institute (Peshawar-Pakistan), vol. 32, no. 1, 2018.

[9] J. Amin, M. Sharif, M. Yasmin, H. Ali, and S. L. Fernandes, “A method for the detection and classification of diabetic
retinopathy using structural predictors of bright lesions,” Journal of Computational Science, vol. 19, pp. 153–164, 2017.

[10] M. D. Abràmoff, P. T. Lavin, M. Birch, N. Shah, and J. C. Folk, “Pivotal trial of an autonomous ai-based diagnostic system
for detection of diabetic retinopathy in primary care offices,” Npj Digital Medicine, vol. 1, no. 1, p. 39, 2018.

[11] Gu, K., Zhai, G., Yang, X., et al.: “A new reduced-reference image quality assessment using structural degradation
model”. IEEE Int. Symp. on Circuitsand Systems (ISCAS), Beijing, China, 2013, pp. 1095–1098

[12] Linquan Lyu; Imad Eddine Toubal; K. Palaniappan, “Multi-Expert Deep Networks for Multi-Disease Detection in Retinal
Fundus Images”, 2022.

Paper id Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org 34