A Dual-Track Framework for Enhancing Age-Related Macular Degeneration Diagnosis Through Embedding and Vision Classification

Age-related macular degeneration (AMD) is a leading cause of vision loss in older adults, making timely and accurate diagnosis crucial for effective treatment. This study suggests a new framework for AMD diagnosis that integrates embedding extraction and direct vision classification to improve diagnostic performance. We employ a dual-track methodology, utilizing advanced pretrained models for embedding extraction (utilizing Virchow2) alongside a vision-based approach (utilizing Nomic) to classify fundus images into distinct AMD categories: (1) normal, (2) intermediate, (3) geographic atrophy, and (4) wet AMD. Our experiments achieved the best results, with an accuracy of 98.03%, an F1 score of 98.05%, and a precision of 98.08%. By experimentally investigating the applicability of pretrained embedding models originally designed for histopathological patches to fundus images, we bridge a gap in multimodal image analysis, potentially paving the way for improved diagnostic tools in ophthalmology. The experimental results demonstrate the potential of combining embedding extraction and direct vision classification to improve the accuracy of AMD detection, addressing the urgent need for automated diagnostic systems in the clinical environment.