Diagnosis of Diabetic Retinopathy Using Deep Neural Network

Abstract

Diabetes-related eye damage is known as diabetic retinopathy. It significantly increases the risk of blindness. Numerous new occurrences of diabetic retinopathy may be decreased with proper eye care. The approach suggested in this study employs U-net segmentation with area fusion and convolutional neural network (CNN) to automatically diagnose and categorize high-resolution retinal fundus pictures into 5 disease stages depending on their severity (Carrera EV, Gonzalez A, Carrera R, Automated detection of diabetic retinopathy using SVM, IEEE XXIV international conference on electronics, electrical engineering and computing (INTERCON), 2017, [1]). When it comes to proliferative diabetic retinopathy, which is characterized by retinal proliferation of neovascularization and retinal detachment, high variability in the categorization of fundus pictures is a significant difficulty (Kumar S, Kumar B, Diabetic retinopathy detection by extracting area and number of microaneurysm from colour fundus image, in 5th international conference on signal processing and integrated networks (SPIN), 2018, [2]). Subsequently, fragmentation of the retina may happen from improper inspection of the retinal vessels, which is required to get an accurate result. Retinal segmentation is a method for autonomously defining blood vessel boundaries. By using region merging, the features are not lost during segmentation and are passed on to the image classifier, which has a 93.33% accuracy rate. Fundus pictures were categorized into no DR, mild, moderate, severe, and proliferative categories based on severity levels. Two datasets were taken into consideration: Diabetic Retinopathy Detection 2015 and Optos 2019 Blindness Detection, both from Cagle. The suggested technique includes data gathering, pre-processing, augmentation, and modeling as its phases. Our suggested model achieves 90% accuracy. Regression analysis was also done, and the accuracy rate was 78%. The main goal of this effort is to provide a trustworthy system for the automated detection of DR.