Explainable Artificial Intelligence for Early Detection and Diagnosis of Age-Related Macular Degeneration

Age-related macular degeneration (AMD) is a leading cause of blindness in older adults. Early identification of individuals at risk of progressing from an asymptomatic stage to advanced AMD is critical for preventing severe vision loss. Automated image assessment systems for AMD can enhance screening efficiency by reducing time, cost, and effort. Despite the success of convolutional neural networks in AMD detection, their limited interpretability restricts their use in clinical practice. To address this issue, we present an explainable deep-learning approach based on ResNetRS-200, incorporating Kernel SHAP for model interpretability. Our approach achieved a quadratic kappa of 0.6927 and an accuracy of 60.95% on the publicly available Age-Related Eye Disease Study dataset, while achieving a quadratic kappa of 0.7784 and an accuracy of 77.95% on a private dataset. Kernel SHAP analysis highlighted specific retinal regions close to the macula influencing the predictions of the model, providing a clinically interpretable framework, and enhancing diagnostic confidence. Our findings demonstrate the effectiveness of the proposed framework for automated AMD grading. Therefore, the proposed method could be an explainable diagnostic aid for the early detection and grading of AMD.Clinical relevanceThis research establishes the useful- ness of an eXplainable Artificial Intelligence approach using ResNetRS-200 architecture and Kernel SHAP for automated AMD grading.