Diabetic Retinopathy Stages Classification using Improved Deep Learning

Abstract

Diabetic Retinopathy (DR) is the most common complication of diabetes mellitus which can cause a loss in vision. The stages of DR can be divided as no DR, non-proliferative DR, and proliferative DR. This paper proposed a method to classify stages of DR using deep learning and genetics algorithm. This research developed an optimal architecture using VGG basic architecture of a convolutional neural network. The results obtained from the Messidor database were 99.66 % accuracy, 99 % sensitivity, and 98 % specificity. Meanwhile, when tested with the Kaggle database the proposed method produced sensitivity, specificity, and accuracy of 98%, 97%, 98.43% respectively. These results show that the method could classify the DR images.