EUVIMEDEUVIMED entdecken
VergleichsstudieOpen Access

Privacy Leakage in Federated Learning in Radiology Reports: Comparative Evaluation of Tokenizer and Batch-Size Privacy Risks

Santhosh Parampottupadam iD, Andrés Martínez Mora iD, Dimitrios Bounias iD, Sinem Sav iD, Klaus Maier-Hein iD, Ralf Floca iD

JMIR Medical Informatics · 11.09.2026 · 14 · e88390-e88390

DOI: 10.2196/88390 ↗

Kurz gefasst

Vergleichsstudie in der Zeitschrift JMIR Medical Informatics, online erschienen am 11.09.2026. Verfasst von Santhosh Parampottupadam und 5 weiteren Autor:innen. Beteiligte Einrichtung: Division of Medical Image Computing, German Cancer Research Center, Im Neuenheimer Feld 280, Heidelberg, Baden-Wurttemberg, 69120, Germany, 49 06221 …. Hauptthemen laut MEDLINE: Federated Learning, Confidentiality, Computer Security, Radiology Information Systems. Die Arbeit nennt 48 Literaturangaben. Der Volltext ist frei in PubMed Central verfügbar.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

Federated learning (FL) enables multi-institutional model training on clinical text without sharing raw data; however, gradient inversion methods can reconstruct sensitive information from shared model updates. The extent of such privacy leakage in FL applied to radiology reports, and the role of tokenizer design, remains unclear.

Objective

This study aimed to quantify gradient-based reconstruction of radiology report text in an FL setting and to compare privacy risk across 3 transformer tokenization strategies in a controlled, tokenizer-aware evaluation.

Methods

Six FL clients trained a GPT-2–style transformer (sequence length 32) on 2 public clinical-text corpora comprising 368,751 diagnostic reports, 98,206 discharge summaries, and 1500 MIMIC-CXR (Medical Information Mart for Intensive Care Chest X-Ray) radiology reports. Models were trained using 3 tokenizers (GPT-2, RadBERT, and LLaMA-2) with batch sizes of 64, 128, and 256. An active malicious-server threat model was assumed, and analytic gradient inversion was applied to recover text. Reconstruction fidelity was measured over 5 runs using exact sentence accuracy, sentence-level bilingual evaluation understudy (S-BLEU), and recall-oriented understudy for gisting evaluation (ROUGE-L).

Results

Exact sentence reconstruction ranged from 27% to 75% across tokenizers, datasets, and batch sizes. At batch size 64 on the discharge dataset, accuracy was 64.7% (GPT-2), 70% (RadBERT), and 67.5% (LLaMA-2), decreasing to 27.3%, 28.5%, and 27.5% at batch size 256. S-BLEU declined with increasing batch size (eg, discharge reports from 0.69 to 0.31). Reconstruction fidelity did not differ significantly across tokenizers (all but one of 27 comparisons nonsignificant; none significant after Holm correction), and approximately 75% of clinical concepts were represented in the reconstructed text (a corpus-level upper bound), regardless of tokenizer. Batch size was the dominant factor governing leakage.

Conclusions

Under a worst-case malicious server that tampers with the shared model and observes unprotected per-client gradients (no secure aggregation or differential privacy), substantial portions of radiology-report text can be reconstructed, with up to approximately 75% of reconstructed 32-token sequences and 75% of clinical concepts (not direct patient identifiers) recovered from FL gradients. In a controlled ablation holding model architecture fixed, tokenizer choice, including domain-specific tokenizers, did not significantly affect leakage under the evaluated conditions, whereas batch size was the primary determinant, and no tokenizer significantly reduced the risk. Tokenizer selection should therefore not be treated as a privacy safeguard in this setting. Safeguards such as secure aggregation and differential privacy should therefore be evaluated as candidate protections for FL deployments that must satisfy Health Insurance Portability and Accountability Act (HIPAA) and General Data Protection Regulation (GDPR) requirements in radiology natural language processing (NLP); legal compliance additionally depends on organizational safeguards, risk assessment, and governance beyond the scope of this study.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Federated LearningConfidentialityComputer SecurityRadiology Information SystemsPrivacyHumans

Stichwörter der Autor:innen

GDPRGeneral Data Protection RegulationHIPAAHealth Insurance Portability and Accountability Actdata securityfederated learninggradient inversionlarge language modelspatient confidentialityprivacyradiologytransformer models

Autor:innen und Institutionen

  • Division of Medical Image Computing, German Cancer Research Center, Im Neuenheimer Feld 280, Heidelberg, Baden-Wurttemberg, 69120, Germany, 49 06221 420.
  • Medical Faculty Heidelberg, Heidelberg University, Heidelberg, Germany.
  • Department of Computer Engineering, Bilkent University, Ankara, Türkiye.
  • Pattern Analysis and Learning Group, Department of Radiation Oncology, Heidelberg University Hospital, Heidelberg, Germany.

ORCID-Profile: Santhosh Parampottupadam, Andrés Martínez Mora, Dimitrios Bounias, Sinem Sav, Klaus Maier-Hein, Ralf Floca

Zitieren

APA

Parampottupadam, S., Martínez Mora, A., Bounias, D., Sav, S., Maier-Hein, K., & Floca, R. (2026). Privacy Leakage in Federated Learning in Radiology Reports: Comparative Evaluation of Tokenizer and Batch-Size Privacy Risks. JMIR Medical Informatics, 14, e88390-e88390. https://doi.org/10.2196/88390

Vancouver

Parampottupadam S, Martínez Mora A, Bounias D, Sav S, Maier-Hein K, Floca R. Privacy Leakage in Federated Learning in Radiology Reports: Comparative Evaluation of Tokenizer and Batch-Size Privacy Risks. JMIR Medical Informatics. 2026;14:e88390-e88390. doi:10.2196/88390

BibTeX
@article{Parampottupadam2026,
  author = {Parampottupadam, Santhosh and Martínez Mora, Andrés and Bounias, Dimitrios and Sav, Sinem and Maier-Hein, Klaus and Floca, Ralf},
  title = {Privacy Leakage in Federated Learning in Radiology Reports: Comparative Evaluation of Tokenizer and Batch-Size Privacy Risks},
  journal = {JMIR Medical Informatics},
  year = {2026},
  volume = {14},
  pages = {e88390--e88390},
  doi = {10.2196/88390}
}
Kostenlos starten →