XV-AMD: An Explainable Vision Transformer Detection Framework for Age-Related Macular Degeneration Using Fundus Imaging

Age-related macular degeneration (AMD) is a leading cause of visual impairment in adults over 50 years of age and is characterized by progressive damage to the macula, leading to challenges in daily activities such as reading and driving. This study introduces the XV-AMD (An Explainable Vision Transformer Detection for Age-Related Macular Degeneration) framework, which uses Vision Transformers (ViTs) and SHapley Additive exPlanations (SHAPs) for the diagnosis and explainability of AMD. Through a systematic preprocessing pipeline, including average cumulative distribution function (CDF) calculation and contrast limited adaptive histogram equalization (CLAHE), our approach enhances image quality and prepares data for analysis. We classify color fundus images into four distinct AMD categories, namely, normal, geographic atrophy (GA), intermediate AMD, and wet AMD, demonstrating the framework’s effectiveness in a clinical context. The experimental results reveal that our ViT-based model not only improves diagnostic accuracy but also facilitates the understanding of decision-making processes, thus supporting early detection and intervention strategies in AMD management. In addition, a 99.54% classification accuracy was attained, with a sensitivity of 99.54%, specificity of 99.85%, precision of 99.54%, and F1 of 99.54%. This research shows the promise of AI and deep learning in improving AMD diagnoses and identifies future paths for boosting model performance and interpretability in medical imaging applications.