Abstract
We analyze the 3D shape of female pelvic organs in two public MRI datasets, using only the segmentation masks, so the remaining signal is purely morphological. Our central finding is a multi-site label trap. On UT-EndoMRI (124 patients from two sites), ovarian and uterine shape predicts whether an endometrioma annotation is present within one site (D2, 15% positive, ROC-AUC 0.826 ± 0.073). However, the prevalence of this annotation differs sharply between sites (78% vs. 15%), so pooled cross-validation is misleading (AUC 0.638), and a model trained at one site fails at the other (AUC 0.274, n=73). Two further results support this caution. First, on UMD (300 women), uterine-corpus shape carries a weak age signal (R2=0.14, MAE 8.9 years) that stems from organ size rather than scale-free shape, while fibroid shape does not predict age at all (R2= − 0.31). Second, scale-free shape descriptors vary less across raters (annotators) than size descriptors (mean coefficient of variation, uterus 0.117 vs. 0.142; ovary 0.147 vs. 0.228). We recommend that shape-based studies in women’s health imaging report label prevalence per site, evaluate cross-site transfer alongside pooled metrics, and quantify the rater robustness of their descriptors. Code is publicly available at https://github.com/TIO-IKIM/shape-demographics-capi
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CAPI_WOMEN_010.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=OIylcbEUcN
BibTex
@InProceedings{LuiGij_AMultiSite_MICCAISAT2026,
author = { Luijten, Gijs AND Engelke, Merlin AND Ruppel, Richard AND Egger, Jan},
title = { { A Multi-Site Label Trap in Shape-Based Endometriosis Detection: What Female Pelvic-Organ Shape Does (and Does Not) Encode } },
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}
}