The abstract is read at the publisher; the summary is JClub's.
The Multimodal Anonymizer achieves 98.80% patient-level deidentification sensitivity on an expanded benchmark of MIMIC-IV-derived bundles, preserving critical clinical content at 99.60%.
- Why it matters: Clinical multimodal data include identifiers in many formats, and existing deidentification approaches need locally deployable, high-sensitivity methods that also retain clinically critical information.
- What they did: The team built a locally deployable framework combining multimodal language models, specialist networks, deterministic transformations, and iterative verification, comparing 16 orchestrator configurations on 250 constructed MIMIC-IV-derived bundles.
- The result: On expanded evaluations including German GRASCCO and 250 Charite partograms, patient-level deidentification sensitivity was 98.80% and per-identifier sensitivity was 99.82%, with all partograms meeting the study criterion.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.