EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Recovering Sensitive Medical Text in Federated Learning

Mohamed El Azzouzi, Reda Bellafqira, Gouenou Coatrieux, Marc Cuggia, Guillaume Bouzille

Studies in Health Technology and Informatics · 2026

DOI: 10.3233/shti260452 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Studies in Health Technology and Informatics, 2026. Verfasst von Mohamed El Azzouzi und 4 weiteren Autor:innen. Beteiligte Einrichtung: Univ Rennes, CHU Rennes, INSERM, LTSI-UMR 1099, F-35000, Rennes, France. Hauptthemen laut MEDLINE: Federated Learning, Electronic Health Records, Confidentiality, Computer Security. Die Arbeit wurde bisher 1-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Federated Learning (FL) allows institutions to train shared models without exchanging raw data, making it a promising approach for healthcare applications that involve sensitive electronic health records (EHRs). However, despite this distributed design, the gradients exchanged during training can still reveal private information. In this study, we analyze how vulnerable transformer-based language models are to gradient inversion attacks, focusing on the Decepticons method, which can reconstruct original training text from shared gradients. We simulate a cross-silo FL setup with three types of French clinical reports (genetic, anesthesia, and birth records) to evaluate how batch size and sequence length affect reconstruction quality. Our experiments show that a malicious server can recover clinical text with high accuracy: token-level recovery exceeded 95% when training with batch size 1 and remained above 60% for sequences of up to 512 tokens. Reconstructed examples contained identifying elements (names, dates, genetic markers), revealing serious privacy risks for real-world use. These results emphasize that FL alone is insufficient for sensitive clinical text and that privacy-preserving defenses must be integrated before real-world deployment.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Federated LearningElectronic Health RecordsConfidentialityComputer SecurityNatural Language ProcessingInformation Storage and RetrievalHumans

Stichwörter der Autor:innen

EHRsFederated LearningGradient LeakageNLPPrivacy

Autor:innen und Institutionen

  • Univ Rennes, CHU Rennes, INSERM, LTSI-UMR 1099, F-35000, Rennes, France.
  • IMT Atlantique, INSERM, LATIM-UMR 1101, F-29200, Brest, France.

Zitieren

APA

El Azzouzi, M., Bellafqira, R., Coatrieux, G., Cuggia, M., & Bouzille, G. (2026). Recovering Sensitive Medical Text in Federated Learning. Studies in Health Technology and Informatics. https://doi.org/10.3233/shti260452

Vancouver

El Azzouzi M, Bellafqira R, Coatrieux G, Cuggia M, Bouzille G. Recovering Sensitive Medical Text in Federated Learning. Studies in Health Technology and Informatics. 2026. doi:10.3233/shti260452

BibTeX
@article{ElAzzouzi2026,
  author = {El Azzouzi, Mohamed and Bellafqira, Reda and Coatrieux, Gouenou and Cuggia, Marc and Bouzille, Guillaume},
  title = {Recovering Sensitive Medical Text in Federated Learning},
  journal = {Studies in Health Technology and Informatics},
  year = {2026},
  doi = {10.3233/shti260452}
}
Kostenlos starten →