Exudate Detection with Improved U-Net Using Fundus Images

Abstract

Diabetic retinopathy (DR) is a chronic disease leading cause of blindness. One of the primary symptoms of DR is exudates (EX). The EX is a condition in which proteins, lipids, water leaked to retinal areas causes vision impairment. The two types of EX are hard EX and soft EX based on their appearance and leakage consistency. Early intervention of DR diminishes the likelihood of vision loss. Therefore, an automated technique is required. We present a novel U-Net model that detects both soft and hard EX in this paper. The proposed model is implemented in two stages. Preprocessing of fundus images is included in the first. The custom residual blocks-based designed network is the second phase. The model is tested on two benchmark databases available publicly IDRiD and e-Ophtha. The results achieved using the proposed approach are better than other approaches.