A Survey on Techniques for Early Detection of Diabetic Retinopathy

Diabetic retinopathy (DR) is a diabetes-related disorder which affects thousands of people worldwide and is one of the major causes of blindness. In this disorder the retina is not able to reflect light that comes in through the eye’s lens properly, thus causing partial of complete blindness. With increasing number of people affected with diabetes, retinopathy is growing to be a serious issue often left uncured until the person goes fully blind; this is in part due to smaller ophthalmologist-to-diabetic patient ratio and the time-consuming process of early detection of retinopathy when done manually. Early detection not only helps the patients with this condition but also is of great assistance to the doctors who will benefit immensely if the process is sped up. Automating this process of early detection can thus help many people from going blind, and with today’s advancement in artificial intelligence techniques especially due to neural networks, it is the best candidate to do this job. In this work we explore several approaches to automating the early detection of diabetic retinopathy using both neural network and traditional computer vision algorithms. The paper compares traditional computer vision-based classification, detection, and segmentation with that of neural network-based approaches and also explores certain works which combine the best of both of these areas for solving computer vision problems. This comparison between various techniques will help in our future work in contributing to the automation of this problem. We have proposed certain neural network architectures which have proven to show state-of-the-art results in related subset of vision problems and which we will explore in further work.