We are delighted to introduce a comprehensive survey examining how federated large language models (LLMs) can support privacy-preserving, trustworthy healthcare systems. Healthcare data is highly sensitive and often siloed across institutions, making it difficult to train powerful LLMs without compromising patient privacy or violating regulatory constraints.
The survey reviews how federated learning allows multiple healthcare providers to collaboratively train LLMs without sharing raw patient data, while examining key applications such as clinical decision support, medical text understanding, and diagnostic assistance. It also maps out open challenges, including communication overhead, model heterogeneity, security vulnerabilities, and fairness across institutions, and proposes future research directions to make federated LLMs more practical and trustworthy for real-world clinical deployment.
The survey was authored by Abderrahman Elhajjout, Zakaria Abou El Houda, Hajar Moudoud, Bouziane Brik, and Mian Ahmad Jan, and has been published in the IEEE Journal of Biomedical and Health Informatics (JBHI).
Warm congratulations to the authors on this important milestone!