DDNet: Diabetic Retinopathy Detection System Using Skip Connection-based Upgraded Feature Block

Diabetic retinopathy is an eye disease that damages the retina caused by diabetes. It affects the eye and eventually impairs vision either completely or partially due to sugar levels. Typically, researchers have been using optical disk segmentation methods to segment diabetic retinopathy images to recognize the severity of the disease on the infected eye. The success of such a technique is heavily dependent on highly skilled and experienced practitioners who have to perform this routine manually and on a case-by-case basis. In this research, we investigate a deep learning methodology for diabetic retinopathy early diagnosis by combining skip connection with upgraded feature blocks using a residual learning strategy. The steps included in the proposed method are data collection, pre-processing, augmentation, and feature modeling. For experimental evaluation, we use a Diabetic Retinopathy Gaussian-filtered Kaggle dataset, which includes Normal, Mild, Moderate, Severe, and Proliferative fundus images. Our proposed approach shows a 3 to 6% improvement over state-of-the-art methods, which illustrates the model's robustness and effectiveness.