Interpretability of Diabetic Retinopathy Images Using Grad-CAM

Abstract

Diabetic retinopathy (DR) is one of the primary reasons for vision loss in people all over the world. The majority of those affected do not have access to expert ophthalmologists or the tools needed to monitor their condition, despite the fact that the disease is quite common. This can cause the start of the treatment to be delayed, which lowers their chances of having a favorable outcome. Deep learning methods that identify the disease in eye fundus photos have been offered as a way to make retinopathy moderate estimations more accessible to patients in remote locations or even as a way to supplement the diagnosis of a human expert in our experiment. In order to serve as a resource for both theorists and practitioners, this paper focuses on deep learning interpretability approach. More specifically, a literature review, and links to Grad-CAM method programming implementations are provided.