A foundation model for generalizable disease detection from retinal images
Problem
AI models for retinal image analysis typically require large labeled datasets and are task-specific, limiting generalizable detection of eye and systemic diseases.
Why it matters
Sight-threatening eye diseases and systemic disorders like heart failure affect millions; earlier detection improves outcomes.
What the AI did
RETFound uses self-supervised learning on 1.6 million unlabelled retinal images to build a foundation model, then adapts it to labeled disease detection tasks.
Results
Adapted RETFound consistently outperformed several comparison models in eye disease diagnosis/prognosis and systemic disorder prediction; no numeric metrics in abstract.
Impact
Peer-reviewed in Nature with 974 citations, indicating strong research uptake.
Record
| Venue | Nature |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Yukun Zhou, Mark A. Chia, Siegfried K. Wagner, Murat Seçkin Ayhan, Dominic J. Williamson, Robbert Struyven, Timing Liu, Moucheng Xu et al. |
| Citations | 975 (OpenAlex, at import time) |
| References | 65 |
| Model or method | RETFound (self-supervised foundation model) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41586-023-06555-x |
| Open access | hybrid |
| Keywords | retinal images, foundation model, self-supervised learning, disease detection, ophthalmology |
Importance rating
major — millions of lives, an industry or a field
Abstract covers diagnosis/prognosis of sight-threatening eye diseases and prediction of systemic disorders like heart failure and myocardial infarction—major health burdens—but states no scale.