Benchmarking Ophthalmology Foundation Models for Clinically Significant Age Macular Degeneration Detection

2025-05-08

AI Summary

This study investigates the generalization capability of Vision Transformers (ViTs) for diagnosing intermediate-to-advanced age-related macular degeneration (AMD) from fundus images, addressing the central question: *Does ophthalmology-specific pretraining outperform generic natural-image pretraining?* We systematically evaluate six self-supervised ViT models across seven multi-center, expert-annotated fundus datasets (70,000 images). Contrary to prevailing assumptions, we find that the natural-image pre-trained iBOT ViT achieves superior cross-distribution AMD diagnosis performance (AUROC: 0.80–0.97), significantly surpassing ophthalmology-specific pretraining (0.78–0.96) and no-pretraining baselines (0.68–0.91). Our contributions include: (1) demonstrating that general-purpose vision foundation models exhibit strong clinical diagnostic generalizability in ophthalmology; (2) releasing BRAMD—the first Brazilian multi-center AMD dataset (n=587); and (3) establishing a standardized ViT evaluation benchmark for cross-center generalization in retinal disease analysis.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multi-instance/Multi-view Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Self-supervised learning (SSL) has enabled Vision Transformers (ViTs) to learn robust representations from large-scale natural image datasets, enhancing their generalization across domains. In retinal imaging, foundation models pretrained on either natural or ophthalmic data have shown promise, but the benefits of in-domain pretraining remain uncertain. To investigate this, we benchmark six SSL-pretrained ViTs on seven digital fundus image (DFI) datasets totaling 70,000 expert-annotated images for the task of moderate-to-late age-related macular degeneration (AMD) identification. Our results show that iBOT pretrained on natural images achieves the highest out-of-distribution generalization, with AUROCs of 0.80-0.97, outperforming domain-specific models, which achieved AUROCs of 0.78-0.96 and a baseline ViT-L with no pretraining, which achieved AUROCs of 0.68-0.91. These findings highlight the value of foundation models in improving AMD identification and challenge the assumption that in-domain pretraining is necessary. Furthermore, we release BRAMD, an open-access dataset (n=587) of DFIs with AMD labels from Brazil.

Problem

• Evaluating ViTs for detecting age-related macular degeneration
• Comparing natural vs ophthalmic pretraining for AMD identification
• Assessing generalization of foundation models on DFI datasets

Innovation

• Self-supervised learning enhances ViTs for AMD detection
• iBOT on natural images outperforms domain-specific models
• Open-access BRAMD dataset supports AMD research