An integrated image processing and deep learning approach for diabetic retinopathy classification

Abstract

Diabetic retinopathy is referred to as eye sight damage and permanent blindness because of diabetic condition in humans. Diabetic patients are growing up in numbers every year around the globe. Because of modern day life style with elevated stress and tension the risk is even higher. Hence it is crucial for diabetic patients to look for its effect on other body parts as diabetic condition hit eye more likely and if left unchecked, may lead to serious eye related issues. But it can be easily monitored by regular checkups and proper health care and prevented from further degradation, while an automatic screening system to identify whether an individual need follow up or referral for supplementary action to avoid blindness can ease the task of detection at early stages to a great extent. In this paper a model for automatic detection of diabetic retinopathy is proposed using low complexity image processing technique and modified Convolutional Neural Network (CNN) with better accuracy and precision to help an ophthalmologist through detection of change in retina features. The proposed model is used to classify the fundus images into two categories, viz. healthy and infected and tested on EyePACS dataset which obtained classification accuracy of 82% shows the robustness of the system.