The Role of Deep Learning in Diagnosis and Grading of Diabetic Retinopathy: A Review

Abstract

In developing nations, diabetic retinopathy is a leading cause of blindness. It often has no early warning signs; however, monitoring the eyes could prevent the progression to more severe forms of sight-threatening retinopathy and macu-lopathy. The primary diagnostic method for diabetic retinopathy is analyzing fundoscopic images; however, manually examining such images can be time-consuming and unreliable because the ability to spot abnormalities varies with the physician's exper-tise. This manual effort has been lessened by computer-aided diagnosis tools that automatically detect diabetic retinopathy on retinal scans. Deep Learning techniques offer an excellent way of creating computer-aided diagnosis tools, delivering highly efficacious diabetic retinopathy detection and grading results. This review lists the current state-of-the-art Deep Learning approaches in the application of detection and grading of diabetic retinopathy, some of which include transfer learning, Convolutional Neural Networks, and object detection techniques, among others. It further identifies potential research gaps in these approaches and the future research direction to find solutions to address them.