AN EFFICIENT METHOD FOR IDENTIFICATION OF SEVERITY LEVEL IN DIABETIC RETINOPATHY USING DEEP NEURAL NETWORKS

Abstract

Listen

Translate article

Diabetic Retinopathy (DR) is an eye-affected diabetes complication. It is caused by disruption to the luminous tissue in the back of the eye's blood vessels. High levels of blood sugar cause blood vessel damage to the retina. The early diagnosis of DR is required because in severe stage it leads to blindness. In this study, an efficient method for DR using deep learning techniques is described. The two tasks are used in this study for performance evaluation. The task 1 is identification of DR using deep learning techniques. Then the task 2 is identifying the severity level. In task 1, initially the Gaussian filter is used for pre-processing to remove noise. Then in Deep Neural Network (DNN), the unimodal features like Visual Geometric Group (VGG) – 16, Xception, NASNET and ResNet V2 are used. The classifiers like Naive Bayes classifier, logistic regression, decision tree, K-Nearest Neighbour (KNN) classifier, Multi Layered Perceptron (MLP), Support Vector Machine (SVM) and DNN. Then in 2nd task the security level is predicted by using DNN. The performance of proposed system is evaluated in experimental results and discussion.