Advancing eye disease detection: A comprehensive study on computer-aided diagnosis with vision transformers and SHAP explainability techniques

Abstract

Eye diseases such as age-related macular degeneration (AMD) and diabetic retinopathy are common worldwide and affect millions of people. These conditions can cause severe vision problems and even lead to blindness if not treated promptly. Therefore, accurate and timely diagnosis is crucial to manage these diseases effectively and prevent irreversible vision loss. This study introduces a computer-aided diagnosis (CAD) framework for automatically detecting various eye diseases via advanced methodologies and datasets. The main focus is on classifying fundus images, which is essential for precise diagnosis and prognosis. By incorporating cutting-edge techniques such as Vision Transformers (ViTs), this study aims to improve the performance and interpretability of traditional Convolutional Neural Networks (CNNs). ViTs can capture complex patterns and long-range dependencies in fundus images, helping distinguish between different eye diseases and healthy conditions. Furthermore, the study integrates SHapley additive exPlanations (SHAP) explainability techniques to provide insights into the model’s decision-making process, enhancing trust and understanding of its predictions. The results demonstrate significant performance enhancements compared with the baseline models, with an overall accuracy of 95%. This method outperforms previous state-of-the-art methods by a considerable margin. Additionally, metrics such as precision, recall, intersection over union (IoU), and the Matthews correlation coefficient (MCC) show superior performance across various eye diseases, such as diabetic retinopathy, glaucoma, and age-related macular degeneration. These findings underscore the effectiveness and reliability of the proposed approach in automated eye disease detection, indicating its potential for clinical integration and widespread adoption in healthcare settings.