Multi-Disease Detection in Retinal Fundus Images Using ViT Architecture

Abstract

Retinal diseases including diabetic retinopathy (DR), age-related macular disease (AMD), and glaucoma are leading causes of vision loss globally. Early detection is critical but limited by the cost and availability of specialized diagnostic tools and expertise, especially in resource-limited settings. To address these challenges, this paper proposes a multi-disease retinal disease classification system using a Vision Transformer (ViT)-based deep learning model combined with a cloud-hosted web application. The system enables healthcare professionals to upload retinal fundus images, receive automated disease classification with confidence scores, and view patient history for longitudinal monitoring. Trained on multiple publicly available datasets, our model achieves classification accuracies above 94% across disease classes and demonstrates robustness to image variability. The lightweight web interface streamlines clinician workflows, enhancing accessibility and timely intervention. This end-to-end solution integrates affordable AI-powered diagnosis with user-friendly cloud services, aiming to democratize retinal healthcare and reduce preventable blindness in underserved populations.