Image Processing Techniques in the Diagnosis of Diabetic Retinopathy: A Research Review

Abstract

Diabetic Retinopathy (DR) is one of the most prevalent and severe microvascular complications associated with diabetes, serving as the leading cause of blindness among adults globally. With the escalating prevalence of diabetes, early detection and timely intervention for DR have become critical public health priorities. However, traditional manual diagnosis of fundus images is hindered by inefficiencies, subjectivity, and reliance on professional expertise, making it inadequate to meet the increasing demands for screening. In recent years, the integration of Artificial Intelligence (AI) and image processing technologies has paved a new path for automated and accurate DR diagnosis. This study systematically reviews the key research advancements and application methods of image processing techniques in DR diagnosis. Initially, it analyzes the role of deep learning models in fundus image analysis across four dimensions: image preprocessing, feature extraction, lesion segmentation, and disease grading. Subsequently, it summarizes the technical characteristics and application outcomes of representative algorithms such as Convolutional Neural Networks (CNNs), U-Net, YOLO, and DarkNet. Finally, it explores current challenges, including dataset imbalance, insufficient model interpretability, and difficulties in clinical implementation. Research findings indicate that deep learning models based on transfer learning and attention mechanisms achieve over 95% sensitivity and above 90% specificity on public datasets, providing robust support for the early screening and precise diagnosis of DR.