Global RETFound: A Protocol for the World's First Global Medical Foundation Model

Foundation models (FMs) are pretrained on large-scale unlabeled data sets and are adaptable to a wide spectrum of downstream diagnostic and prognostic tasks. However, limited representation in existing ophthalmic artificial intelligence (AI) data sets, particularly from Latin America, Africa, the Middle East, and Central Asia, could exacerbate health disparities. We describe an international collaborative protocol to develop and validate Global RETFound, an open-access FM for ophthalmic and oculomics research, pretrained on synthetic and anonymized real color fundus photographs (CFPs).

International multicenter study.

Data from a global consortium of institutions.

International collaborators are supported with Docker-based tools, model/data sharing agreements, open-access licenses, tailored protocols, and standardized guides. Under a flexible 2-tier contribution framework, collaborators share fine-tuned generative AI (GenAI) weights or anonymized real CFPs.

Prespecified outcomes include successful development of a globally representative, open-access FM and downstream performance evaluation across ocular and systemic diseases, assessed using discrimination metrics (area under the receiver operating characteristic curve and area under the precision-recall curve), sensitivity, specificity, F1-score, calibration, and task-specific regression and segmentation measures.

The Global RETFound consortium has established an international collaborative study with broad geographic representation. As of the 10th of July 2026, 232 collaborators from 78 countries have been onboarded, and 88 study assets (in the form of GenAI weights or anonymized real CFPs) from 39 countries have been shared.

The Global RETFound protocol is a novel, privacy-preserving, inclusive strategy for developing a large-scale FM using synthetic and real CFPs. By enabling global participation from institutions with varying technical resources, and focusing on underrepresented populations, the initiative aims to advance equitable and generalizable FMs for ophthalmic and systemic disease. This initiative is positioned as a blueprint for future global FM collaborations in other medical specialties.

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.