DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING

Domain Name : MEDICAL IMAGE PROCESSING

CLASSIFICATION OF DIABETIC RETINOPATHY USING DEEP

LEARNING WITH MATLAB
Team Members
Under the Esteemed Guidance of G.BHANU PRAKASH - 216M1A0434
Ms. R. SMILE, M.Tech D.PRANEETH - 216M1A0421
Assistant Professor K. UDAY SAI RAJ - 216M1A0457
G.VINOD
- 216M1A0437
TABLE OF CONTENTS
 Introduction
 Abstract
 Objective
 Literature Survey
 Existing System
 Proposed System
 Results
 Module Description
 Future Scope
 Conclusion
 Work Done
 Road Map
 References
 INTRODUCTION
IMAGE PROCESSING
Image processing is the manipulation and analysis of digital
images using a computer to:
 Improve image quality.
 Extract useful information.
 Recognize patterns.
 Enhance visual representation.
 MEDICAL IMAGE PROCESSING

Medical imaging is the

technique and process of
imaging the interior of a body
for clinical analysis and
medical intervention, as well
as visual representation of the
function of some organs or
tissues .
 FUNDUS PHOTOGRAPHY
Fundus photography is a non-invasive imaging technique used to
capture a picture of the back of the eye, known as the fundus.

Types of Fundus Photography:

Types of Fundus Photography:
•Color Fundus Photography
 Color Fundus Photography.
 Red-Free Fundus Photography.
 Fluorescein Angiography.
 Optical Coherence Tomography.
 HUMAN EYE STRUCTURE

Types of Fundus Photography:

•Color Fundus Photography
 DIABETES
 Diabetes is a chronic health condition that affects how your body turns
food into energy. Essentially, it's a problem with insulin, a hormone
produced by your pancreas.
Types of Diabetes:
There are two main types of diabetes:
 Type 1 diabetes: Your body's immune system attacks and destroys the
cells in your pancreas that make insulin. This means your body makes
little or no insulin.
 Type 2 diabetes: Your body becomes resistant to insulin, and your
pancreas may not be able to produce enough insulin to overcome this
resistance.
 DIABET
HIGH BLOOD PRESSURE ES GENETICAL DISORDER
RISK
FACTOR
S

EYE SIGHT
OVER WEIGHT LACK OF ACTIVITY
 TREARMENT FOR DIABETIC RETENOPATHY

 Anti-VEGF Injections
 Steroid Injections
Anti-VEGF Injections
 Focal Laser Treatment
 Panretinal Photocoagulation
 Vitrectomy
 DATA SET AVAILABILITY
MODALITY FUNDUS PHOTOGRAPHY

DESCRIPTION Contains FUNDUS Photography images

classified into Mild stage, Moderate
stage, No DR, Proliferative Stage, Severe
Stage.
SOURCE https://www.kaggle.com/datasets/
amanneo/diabetic-retinopathy-resized-
arranged
NO.OF SUBJECTS Total - 3500/ MILD-700/MODERATE-
700/NO DR-700/PROLIFERATIVE-
700/SEVERE-700
USE CASE CLASSIFICATION TASK
 ABSTRACT
This research investigates the application of deep learning for automated
diabetic retinopathy (DR) detection using retinal fundus images, aiming
to provide a rapid and reliable screening tool. We employ convolutional
neural networks (CNNs), including transfer learning strategies, to classify
images into distinct DR stages, addressing the challenges of data
imbalance and limited generalization. The proposed system is evaluated
using metrics such as accuracy, sensitivity, and specificity, demonstrating
the potential of deep learning to enhance early DR detection and facilitate
timely intervention, ultimately contributing to the prevention of vision
loss.
 OBJECTIVE

To develop a highly accurate and

automated deep learning system
capable of classifying retinal fundus
Anti-VEGF Injections
images into various stages of
diabetic retinopathy, enabling early
detection and timely intervention.
 LITERATURE SURVEY
AUTHOR TITLE MODALITY DATABASE ALGORITHM FINDINGS

Deep Learning FUNDUS Kaggle CNN Detection of

Approach to Diabetic Diabetic
Borys Tymchenko Retinopathy Detection Retinopathy with
92% accuracy

Gazala Mushtaq Detection of diabetic Anti-VEGF

FUNDUS Injections
Kaggle VGG16 Achieved 90%
retinopathy using deep accuracy in
learning detecting the DR.
methodology
Zubair Khan Diabetic Retinopathy FUNDUS KAGGLE VGG-NIN Reached 94%
Detection Using VGG- accuracy for
NIN Diabetic
a Deep Learning Retinopathy
Architecture detection with pre-
trained models
 AUTHOR TITLE MODALITY DATABASE ALGORITHM FINDINGS

A Deep Learning FUNDUS Kaggle CNN Detection of

Ensemble Approach Diabetic
Sehrish Qummar for Diabetic Retinopathy with
Retinopathy 92% accuracy
Detection
Chun‑Ling Lin1 Development of FUNDUS Kaggle RESENET Achieved 96%
revised ResNet‑50 accuracy in
Anti-VEGF Injections
for diabetic detecting the DR.
retinopathy detection
Ayesha Jabbar A Lesion-Based FUNDUS KAGGLE CNN Reached 94%
Diabetic Retinopathy accuracy for
Detection Through Diabetic
Hybrid Deep Retinopathy
Learning Model detection with pre-
trained models
 AUTHOR TITLE MODALITY DATABASE ALGORITHM FINDINGS

Diabetic Retinopathy FUNDUS Kaggle CNN Detection of

Detection via Deep Diabetic
Convolutional Retinopathy with
Zhiguang Wang Networks for 96% accuracy
Discriminative
Localization and
Visual Explanation
Javeria Amin, A Review on Recent Anti-VEGF
FUNDUS Injections
Kaggle VGG16 Achieved 90%
Developments for accuracy in
Detection of Diabetic detecting the DR.
Retinopathy
Imran Qureshi Recent Development FUNDUS KAGGLE CAD Reached 92%
on Detection Methods accuracy for
for the Diagnosis of Diabetic
Diabetic Retinopathy Retinopathy
detection with pre-
trained models
 EXISTING SYSTEM
 Machine learning algorithms are used to classify images into different DR
stages. Common classifiers include:
a) Support Vector Machines (SVMs)
b) Random Forests
c) k-Nearest Neighbors (k-NN) Anti-VEGF Injections
d) Naive Bayes
 Advanced models, especially those using transformers or high-resolution
images, can be computationally intensive, requiring powerful hardware.
 Although there are efforts to create real time analysis, some systems still
require the images to be uploaded to a cloud-based service, which can
cause delays.
 PROPOSED SYSTEM
 Leveraging Convolutional Neural Networks (CNNs) for efficient and
automated analysis of retinal fundus images.
 Reducing the need for time-consuming manual grading by
ophthalmologists.
 Applying image preprocessing
Anti-VEGF Injections
steps (e.g., contrast enhancement,
noise reduction) to improve image quality and model performance.
 Implementing robust data augmentation techniques to enhance
dataset variability and address data imbalance.
 Conducting rigorous cross-validation to ensure model generalization.
 RESULTS
MODEL USED TRAINING ACCURACY TESTING ACCURACY

CNN 10 93.20 94.43

SGD 20.00Injections
Anti-VEGF 23.00

ADAGRAD 81.80 82.33

EfficientNet-B0 81.80 83.64

RESENET 50 94.84 95.55

CNN 10

Anti-VEGF Injections
Anti-VEGF Injections
SGD OPTIMIZATION

Anti-VEGF Injections
Anti-VEGF Injections
ADAGRAD OPTIMIZATION

Anti-VEGF Injections
Anti-VEGF Injections
EfficientNet-B0

Anti-VEGF Injections
Anti-VEGF Injections
RESENET 50

Anti-VEGF Injections
Anti-VEGF Injections
 MODEL DESCRIPTION
 RESENET-50 is a pre-trained deep learning model designed for image classification and
detection. It is faster and more accurate than other models.
 Input Size – Takes 224×224 RGB images as input.
 Optimized Deep Learning Model – Uses compound scaling to balance depth, width, and
resolution.
 Layers – ResNet-50 has 50 layers, including convolutional layers, batch normalization
layers, ReLU activation functions, and pooling layers..
 Higher Accuracy – Outperforms ResNet-50, VGG16, DenseNet-201 in detection tasks.
 FUTURE SCOPE
The future scope of diabetic retinopathy (DR) detection using deep learning is vast and
holds immense potential for improving patient outcomes.
Personalized DR Management:
 Developing CNN-based systems that can predict the progression of DR based on
individual patient data, enabling personalized treatment plans.
 Combining CNN analysis with other patient data (e.g., blood sugar levels, genetic
information) for more comprehensive risk assessment.
Multi-Modal Imaging Analysis:
 Developing CNNs that can analyze data from multiple imaging modalities (e.g., fundus
photography, OCT, OCTA) to provide a more comprehensive view of retinal health.
 Improving the accuracy and sensitivity of DR detection by leveraging complementary
information from different imaging techniques.
 CONCLUSION

 This project successfully implemented Diabetic Retinopathy detection using deep

learning in MATLAB. Various pretrained models, including ResNet-50, CNN-10
EfficientNet-B0, ResNet-50, SGD OPTIMIZATION, ADAGRAD OPTIMIZATION,
CNN-10 were analysed for performance. ResNet-50 outperformed other models,
proving to be the most effective for accurate classification. The proposed model
successfully implemented Diabetic Retinopathy detection with 94.10% accuracy by
reducing manual effort, human error.
 WORK
DONE
 Researched about Diabetes, Diabetes classifications and detection.
 Confirmed the Appropriate Title. Collected data sets from Kaggle - 3500
 Balanced the data sets in each field to 700.
 Model Selection – Used multiple deep learning models (EfficientNet-B0, ResNet-50,

SGD OPTIMIZATION, ADAGRAD OPTIMIZATION, CNN 10).

 Training & Testing – Trained models using MATLAB, evaluated performance using
accuracy, confusion matrix, and ROC curve.
 Best Model Identified – CNN 10 achieved the highest accuracy among all tested
models.
 ROAD MAP

Data Set collection Algorithm

Report
and Literature Implementation
Submission
survey Results
Anti-VEGF Injections

REVIEW - 1 REVIEW - 2 REVIEW - 3

10/04/2025 34
 REFERENCES
1. https://cdn.aaai.org/ocs/1186/1186-5878-1-PB.pdf
2. https://ieeexplore.ieee.org/ielx8/6287639/10380310/10559800.pdf?tp=&arnumber=10
559800&isnumber=10380310&ref=aHR0cHM6Ly9pZWVleHBsb3JlLmllZWUub3Jn
L2Fic3RyYWN0L2RvY3VtZW50LzEwNTU5ODAw
3. https://ieeexplore.ieee.org/ielx7/6287639/10380310/10459189.pdf?tp=&arnumber=10
459189&isnumber=10380310&ref=aHR0cHM6Ly9pZWVleHBsb3JlLmllZWUub3Jn
L2Fic3RyYWN0L2RvY3VtZW50LzEwNDU5MTg5
4. https://iopscience.iop.org/article/10.1088/1757-899X/1070/1/012049/pdf
5. https://www.researchgate.net/publication/370305710/figure/fig1/AS:11431281153899
905@1682599722777/Schematic-diagram-of-the-68-facial-landmarks-The-photo-use
d-for-the-schematic-diagram-is.png

https://spj.science.org/cms/10.34133/2021/8786793/asset/aa5477a0-6fcf-4a76-bd5c-b
dcd79ce919f/assets/graphic/8786793.fig.001.jpg
6. https://mohaneyeinstitute.com/wp-content/uploads/2022/08/Diabetic-Retinopathy-
iStock-695338058-1030x703.jpg
7. https://cdn.aaai.org/ocs/ws/ws0448/16668-75987-1-PB.pdf
8. https://ieeexplore.ieee.org/ielx7/6287639/8600701/08869883.pdf?tp=&arnumber=
8869883&isnumber=8600701&ref=aHR0cHM6Ly9pZWVleHBsb3JlLmllZWUub
3JnL2Fic3RyYWN0L2RvY3VtZW50Lzg4Njk4ODM=
9. https://link.springer.com/content/pdf/10.1007/s40123-018-0153-7.pdf
10. https://mdpi-res.com/d_attachment/symmetry/symmetry-11-00749/article_deploy/
symmetry-11-00749.pdf?version=1559563052
11. https://link.springer.com/content/pdf/10.1186/s12859-023-05293-1.pdf
12. https://iopscience.iop.org/article/10.1088/1757-899X/1070/1/012049/meta
13. https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9409084
14. https://arxiv.org/pdf/2003.02261
15. https://www.aarp.org/content/dam/aarp/health/conditions_treatments/2020/11/114
0-eye-anatomy-diagram.web.jpg
THANK YOU