ummadiviany / retinal-diabetic-retinopathy-grading

This repository is mainted for term project Retinal Diabetoc Retinopathy Grading for Medical Image Analsysis class taught by Prof Nirmalya Ghosh at IIT Kharagpur.

retinal-diabetic-retinopathy-grading

This repository is mainted for term project Retinal Diabetoc Retinopathy Grading for Medical Image Analsysis class taught by Prof Nirmalya Ghosh at IIT Kharagpur.

This page is under active development

The project report can be found here : Project Report

The project slides can be found here : Project Slides

Problem

Diabetic Retinopathy is a complication of diabeties that effect eye. DR is the major cause of blindness in India, which accounts for 30% of DR cases in the world. Early diagnosis of DR can reduce the risk of blindness by 90%. Causes for DR can be identified using the following features:

1. Hemmorages
2. Abnormal growth of blood vessels
3. Exudates

Now lets see how DR can be classified into different types. The following table shows the classification of DR.

Class	No DR	Mild DR	Moderate DR	Severe DR	Proliferate DR
Image
Description	No presence of DR features. Clean Eye	At least one microaneurysm present on retinal exam	Multiple microaneurysms, dot-and-blot hemorrhages, venous beading, and/or cotton wool spots	Cotton wool spots, venous beading, and severe intraretinal microvascular abnormalities (IRMA)	Growth of new blood vessels, Blood vessels bleeding, Retinal detachment

Objective

The objective our work is to classify the DR into different types. This is called as DR grading. The grading labels are as follows:

0. No DR 
1. Mild DR
2. Moderate DR
3. Severe DR
4. Proliferate DR

We would like to classify the input image into one of those five classes.

Methods

Binary Classification

First we will consider a simpler task of binary classification of DR. Its basically saying DR or No DR. To achieve this, the four DR classes are merged into meta class DR. Now its a binary classification problem. Deep neural netoworks can be used to classify the image into one of the two classes. Four standard deep neural network architectures are used to classify the image. The performance results are shown below table.

Network	Top 1 Accuracy	Precision	Recall	F1 Score	False classifications(out of 800)
VGG19	0.977	0.98	0.98	0.98	18
ResNet18	0.974	0.97	0.97	0.97	21
EfficientNetB0	0.974	0.97	0.97	0.97	21
ConvNext-Tiny	0.981	0.98	0.98	0.98	13

ConvNext-Tiny is the best performing model. It has a top 1 accuracy of 0.981 and a precision of 0.98, recall of 0.98 and F1 score of 0.98. It has a false classifications of 13.

Multi Class Classification

Network	Top 1 Accuracy	Precision	Recall	F1 Score	False classifications(out of 800)
VGG19	0.9644	0.96	0.96	0.96	25
ResNet18	0.99	0.98	0.98	0.98	7
EfficientNetB0	0.997	0.99	0.99	0.99	2
ConvNext-Tiny	0.995	0.99	0.99	0.99	3

EfficientNetB0 is the best performing model. It has a top 1 accuracy of 0.997 and a precision of 0.99, recall of 0.99 and F1 score of 0.99. It has a false classifications of 2.