A Vision Transformer-Based Convolutional Neural Network for the Automated Diagnosis of Eye Diseases Using Self-Attention Mechanisms

Abstract

Daily life is highly dependent on the eyes, making them one of the most essential organs in the body. This study focuses on four eye conditions: Normal, Diabetic Retinopathy, Cataracts, and Glaucoma. This study presents a Convolutional Neural Network (CNN) model based on a Vision Transformer (ViT) with a Self-Attention Mechanism (SAM) for diagnosing various eye diseases. Initially, the dataset was preprocessed through resizing and normalization to enhance image quality and facilitate feature extraction. The proposed model was evaluated, achieving a commendable accuracy of 94% on test data, with an average AUC of 98.82%. This model effectively diagnoses conditions such as Diabetic Retinopathy, Cataracts, Glaucoma, and normal cases. The GUI-based application was developed and tested, allowing doctors to upload multiple images and analyze eye disease categories, enhancing interpretability and showing promise for clinical applications. The proposed model can assist ophthalmologists in detecting eye disorders, enabling timely treatment of patients and helping to prevent vision loss.