DR-Net: CNN Model to Automate Diabetic Retinopathy Stage Diagnosis

Abstract

Diabetic Retinopathy (DR) is an eye-related disease occurs in diabetic patients due to a rise in blood sugar level. As the diabetes advances in stage, patients eyesight may weaken that is the sign of the early stage of DR. On increasing blood sugar in diabetic patients, DR becomes a major concern of the world's population as the advance stage cause complete vision loss. The early detection is necessary for the treatment, but the diagnosis of DR is difficult and costly. The task demands expert clinicians to find out the existence of different features present at different stages of DR, which is time-consuming. In this paper, a CNN model (DR-Net) is proposed that automate the detection of DR and its severity in retinal images. Retinal fundus images are taken from the available dataset and are used to train, validate the network and test the accuracy of the proposed method. The experimental result shows that this CNN based classifier is able to classify the retinal images into five stages of the disease namely No DR, Mild DR, Moderate DR, Severe DR and Proliferative DR based on severity. The model achieved good results as compared to existing classifier model for four-classes.