Deep learning for early detection and classification of diabetic retinopathy using fundus images

BackgroundDiabetic retinopathy (DR) is a critical complication of diabetes, leading to global vision impairment. Early detection and precise classification of DR are essential for effective intervention and improved patient care. However, early-stage DR often shows few symptoms, making early identification and treatment challenging.ObjectiveManual diagnosis of DR from fundus images is time-consuming, costly, and prone to misdiagnosis compared to computer-aided diagnosis systems.MethodsThis paper presents the development and evaluation of advanced deep learning models for the detection and classification of DR using both binary and multiclass classification approaches.ResultsWe employed deep learning techniques on fundus image datasets to perform both 2-class and 5-class classifications. The models were trained and tested with the support of ophthalmologist surgeons at the Eye Hospital of Biskra, Algeria.ConclusionsThe results demonstrated significant efficacy, garnering positive feedback from hospital physicians. Our proposed model achieves 98.7% accuracy for 2-class classification and 96.2% accuracy for 5-class classification in recognizing and classifying the severity levels of DR, including no DR, mild DR, moderate DR, severe DR and proliferative DR.