- Preprint Article
- 10.36227/techrxiv.177203265.53959182/v1
Federated Learning for Ophthalmic Imaging: A Systematic Review
- Feb 25, 2026
- Luyuan Qi + 5 more +5
Ophthalmic diseases are leading causes of vision loss worldwide, and medical imaging-based AI models are increasingly used to support their detection and management. However, the development of robust and generalisable models is hindered by siloed imaging data across institutions. Federated learning (FL), which is a collaborative machine learning approach that trains models across multiple decentralised devices or institutions while keeping all data local to preserve privacy, has emerged as a promising solution. This systematic review synthesizes current research on FL for ophthalmic disease detection, focusing on methodological trends, privacy-preserving techniques, and challenges in clinical translation. A total of 22 peerreviewed studies were identified. Across classification and segmentation tasks involving fundus, OCT, and OCT-A images, FL generally achieved performance comparable to centralised training and superior to local models. Privacy preservation has primarily relied on differential privacy, with growing exploration of cryptographic and blockchainbased strategies. Domain heterogeneity remains a key technical challenge, with most studies relying on simulated client splits that fail to fully capture real-world variability. Overall, current ophthalmic FL research is in its early stages, with promising results but limited real-world validation. Future work should prioritize clinically realistic multi-institutional datasets, stronger privacy-utility tradeoffs, robust domain generalisation, and clear pathways to regulatory adoption.
Read more
1
