Abstract
Radiology archives accumulate two kinds of identity error: the same study stored twice under different identifiers, and a study filed under the wrong patient. Both are easy to catch when the two images use the same MRI sequence, but not when they differ—a T1-weighted and a T2-weighted scan of the same pelvis share almost no pixel-level appearance. We ask whether frozen foundation-model embeddings can decide, from the image alone and without metadata, whether two pelvic MRI series come from the same patient, even across sequences. Across 1,087 volumes from 252 women in six cohorts, we encode each volume into a single vector with a frozen 2D backbone and compare vectors with a small learned similarity. Matching across sequences reaches 0.89 AUROC, as reliable as matching within the same sequence, while plain embedding distance (cosine) recovers only a weak signal (0.66). Occlusion and region analyses place the identity signal in the bony pelvis, and it persists across imaging timepoints. The signal does not transfer between institutions: trained on other sites, cross-sequence AUROC falls to about 0.6. Image-based cross-sequence identity is therefore a practical same-site quality-assurance check and a concrete within-site privacy risk.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CAPI_WOMEN_014.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/CAPI_WOMEN_014_supp.zip
Link to Open Review
Open Review Page: https://openreview.net/forum?id=xCfJ5stIWP
BibTex
@InProceedings{RupRic_Pelvic_MICCAISAT2026,
author = { Ruppel, Richard AND Lindholz, Maximilian AND Hamm, Charlie Alexander AND Jeanette Savic, Lynn AND Schmidt, Robin AND Arlt, Tillmann AND Luijten, Gijs AND Stepansky, Leonard AND Bauer, Daniel AND May, Matthias Stefan AND Penzkofer, Tobias},
title = { { Pelvic MRI Patient Verification with Foundation-Model Embeddings } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17256},
month = {pending},
page = {pending}
}