Abstract

The rise of electronic health records (EHRs) has unlocked new opportunitiesfor medical research, but privacy regulations and data heterogeneity remain keybarriers to large-scale machine learning. Federated learning (FL) enablescollaborative modeling without sharing raw data, yet faces challenges inharmonizing diverse clinical datasets. This paper presents a two-step dataalignment strategy integrating ontologies and large language models (LLMs) tosupport secure, privacy-preserving FL in healthcare, demonstrating itseffectiveness in a real-world project involving semantic mapping of EHR data.

 

Quick Read (beta)