Beyond the foundation model: A practical framework for adaptation in ophthalmology.

Authors

Ferdinando Giuliano, Yalin Zheng, Enrico Giuliano, Vito Romano, Mariantonia Ferrara, Mario R Romano

Published in

Acta ophthalmologica. Nov 14, 2026. Epub Nov 14, 2026.

Abstract

Foundation models are rapidly emerging as a major development in ophthalmic artificial intelligence, largely because they can achieve strong downstream performance with fewer labelled images than conventional supervised models, including those pretrained on natural images, although the size of this advantage varies across tasks. While recent studies have described the architectures and benchmark results of ophthalmic foundation models, less attention has been paid to a more practical question: Once such a model is available, how should a clinician or researcher adapt it to a specific task? This review addresses this question from a practical perspective. We first discuss what ophthalmic foundation models can learn during pretraining and how these learned representations can be transferred across downstream tasks. We then examine the main adaptation strategies that are currently used in practice, with attention to their relative strengths, limitations and data requirements. We also review key methodological considerations that influence whether adaptation is likely to be reliable, including annotation quality, dataset construction, patient-level splitting, external validation, calibration and explainability. The main aim of this review was not to categorize foundation models themselves, but to provide a clinically oriented framework for deciding when fine-tuning is likely to be useful and when simpler alternatives may be sufficient. We argue that this decision should weigh labelled-data availability, disease rarity, domain match between pretraining and target task, model availability and licensing, computational constraints and the need for generalization across devices or populations, none of which alone justifies fine-tuning, and that any adapted model should be compared with a supervised baseline and validated in a manner matched to its intended use. By shifting the focus from model development to model adaptation, this review aimed to support more informed and realistic use of foundation models in ophthalmic clinic and research.

PMID:
43031527
Bibliographic data and abstract were imported from PubMed on 14 Nov 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.