Computer-Aided Multi-Class Classification of Diabetic Retinopathy Using Fundus Images

Abstract

Diabetic Retinopathy is a progressive retinal disease and a major cause of preventable vision loss in those with diabetes. For immediate and appropriate clinical interventions in DR, recognising and appropriately understanding its severity are quite important. However, manual classification of diabetic retinal images and their severity assessment are time-consuming and highly dependent on specialists. Therefore, in this research, a multi-class classification system has been presented for automated classification of retinal images from people with diabetes and the identification of their severity levels. The proposed technique uses deep learning feature generation alongside supervised learning to identify and predict distinct visual features and varying degrees of DR. Traditional image pre-processing is performed to improve image quality and reduce light intensity variations. Moreover, a large number of data augmentation techniques are employed to improve the classifier's generalisation capacity