Privacy Leakage in Federated Learning in Radiology Reports: Comparative Evaluation of Tokenizer and Batch-Size Privacy Risks - Takeaways - MDSpire
Coming Soon: Introducing MDSpire News. Learn more
Conexiant’s news site is now MDSpire News. Learn more

Evaluating Privacy Risks in Federated Learning for Radiology Reports: A Comparison of Tokenizer and Batch Size Vulnerabilities

  • By

  • Santhosh Parampottupadam

  • Andrés Martínez Mora

  • Dimitrios Bounias

  • Sinem Sav

  • Klaus Maier-Hein

  • Ralf Floca

  • September 11, 2026

Share

  • 1

    Radiology reports provide critical clinical information that enhances medical diagnostics beyond imaging data.

  • 2

    Federated learning (FL) allows collaborative model development while retaining sensitive data at individual institutions.

  • 3

    FL systems can transmit model parameters, which may expose vulnerabilities that threaten patient confidentiality.

  • 4

    Gradient-inversion attacks can potentially reconstruct sensitive data from model parameters shared in FL systems.

  • 5

    Domain-specific tokenizers and LLM training on specialized corpora can enhance performance in medical NLP applications.

Original Source(s)

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

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

  • September 11, 2026

Related Content

Educators weigh how to use AI without losing critical thinking

Greater productivity is not necessarily indicative of greater knowledge.

Signal, not noise: learning to measure emphysema beyond the −950 HU threshold

Prognostic value of baseline PSMA PET/CT for survival in synchronous metastatic hormone-sensitive prostate cancer: towards redefining low- and high-volume disease