Deep Learning Model to Predict the Risk of Developing Diabetic Retinopathy

Abstract

that causes diabetic patients to go blind due to the damage of retinal blood vessels. Initially it is asymptotic, but it affects both the eyes and eventually causes partial or complete vision loss if it becomes severe. The most effective strategy to manage the condition is to have regular fundus photography screenings and timely management. The increased number of diabetic patients and their extensive screening needs have sparked interest in a computer-assisted, totally automatic diagnosis of DR.  The early detection of DR can save the diabetic people from permanent blindness. The goal is to design a deep-learning system, specifically an Inception-v3 model, that could predict the probability of diabetic retinopathy developing within two years in patients with diabetes. The present work is developed and tested on two versions of a deep-learning system to predict the progression of diabetic retinopathy in diabetic patients who have undergone tele-retinal diabetic retinopathy screening in a primary care environment. A risk categorization technique like this could help to improve screening intervals while lowering costs and increasing vision-related outcomes.

Keywords: Diabetic Retinopathy, Inception-v3, deep learning, Gabor filter, fundus images.