Vision Transformer and Flask-Based Web System for Diabetic Retinopathy Detection

One of the main causes of vision loss worldwide, especially in people with long-term diabetes, is diabetic retinopathy (DR). Timely intervention and early detection are essential for avoiding blindness or severe visual impairment brought on by this illness. Traditional diagnostic techniques, however, mostly depend on skilled ophthalmologists manually examining retinal fundus images, which takes a lot of time and money. Many rural or underprivileged locations lack access to qualified medical personnel and diagnostic equipment, which causes missing or delayed diagnosis. For DR stage detection, an automated, scalable, and precise solution is essential. While guaranteeing prompt identification of patients who are at danger, such a method might greatly lessen the diagnostic load on medical personnel. By adding unique classification layers, such as a global average pooling layer, dropout for regularization, and a dense layer with sigmoid activation for multilabel classification, the Vision Transformer (ViT) model—more especially, the vit base patch16 224 architecture—is optimized for DR detection. Using its self-attention mechanism, this transformer-based method is excellent at capturing global dependencies and tiny fluctuations in retinal pictures. This is especially useful for identifying subtle traits that correspond to various phases of DR. Through the use of cutting-edge deep learning technologies like ViT and their deployment via intuitive platforms like Flask applications, this system improves diagnostic precision, democratizes access to DR screening, and helps medical professionals handle the rising incidence of diabetes-related eye disorders.