A review published in the IEEE Journal of Biomedical and Health Informatics in 2026 examines ophthalmic foundation models, which apply AI to eye care. The review says recent advances have improved disease detection, progression assessment and treatment evaluation. It attributes progress to greater availability of multimodal eye-imaging data, including optical coherence tomography, color fundus photography and slit-lamp imaging, as well as self-supervised learning, vision-language modeling and agent-based systems.

The authors summarize publicly available multimodal ophthalmic imaging datasets and their role in large-scale pretraining and cross-modal representation learning. They group the models into vision foundation models, vision-language foundation models and large language model-based agentic systems, and discuss their architectures, learning strategies and applications. Across structural, vascular and anterior-segment imaging, the review describes a move from task-specific image recognition toward clinically oriented systems with semantic grounding and interactive reasoning.

The review identifies barriers to clinical translation: imbalanced data across modalities, imperfect alignment between images and text, limited robustness and generalizability, hallucinations by generative agents, and difficulty integrating systems into clinical workflows. It recommends standardized multimodal data resources, better interpretability and factual consistency, rigorous multi-center validation and cross-disciplinary collaboration to support reliable clinical decisions.

A review published in the IEEE Journal of Biomedical and Health Informatics in 2026 examines ophthalmic foundation models, which apply AI to eye care. The review says recent advances have improved disease detection, progression assessment and treatment evaluation. It attributes progress to greater availability of multimodal eye-imaging data, including optical coherence tomography, color fundus photography and slit-lamp imaging, as well as self-supervised learning, vision-language modeling and agent-based systems.

The authors summarize publicly available multimodal ophthalmic imaging datasets and their role in large-scale pretraining and cross-modal representation learning. They group the models into vision foundation models, vision-language foundation models and large language model-based agentic systems, and discuss their architectures, learning strategies and applications. Across structural, vascular and anterior-segment imaging, the review describes a move from task-specific image recognition toward clinically oriented systems with semantic grounding and interactive reasoning.

The review identifies barriers to clinical translation: imbalanced data across modalities, imperfect alignment between images and text, limited robustness and generalizability, hallucinations by generative agents, and difficulty integrating systems into clinical workflows. It recommends standardized multimodal data resources, better interpretability and factual consistency, rigorous multi-center validation and cross-disciplinary collaboration to support reliable clinical decisions.