Abstract
Patient-specic 3D hip morphology remains dicult to generate at clinical scale because hospital imaging archives are rarely linked to governed segmentation and shape-analysis pipelines. We present a proof-of-concept hospital Trusted Research Environment workow that connects PACS-to-XNAT transfer and JupyterHub/DGX-A100 compute to a two-phase analysis. Phase 1 segments acetabulum, femur and acetabular labrum with MedSAM2-Anatomy, a training-free inference-time optimisation that takes localisation from a frozen expert and delineation from a frozen promptable foundation model: TotalSegmentator for bone and a supervised nnU-Net (HipLabrum) for the labrum. Phase 2 runs two independent pipelines on the accepted masks of both cohorts, classical morphometrics and particle-based shape models, and correlates both separately with clinically documented femoral torsion. We asked whether the workow yields segmentations reliable enough for 3D analysis and whether the resulting features track torsion, separately in femoroacetabular impingement (F AI) and hip dysplasia (HD). On the 12 expertannotated hips, MedSAM2-Anatomy reached mean Dice 0.886 (acetabulum) and 0.890 (femur) against 0.501 and 0.657 for the frozen prior alone, and lowered median HD95 to 3.67 and 2.31 mm. For the labrum the supervised localiser stayed ahead (0.635 vs. 0.620), so the contribution there is multi-anatomy integration, quality assurance and provenance rather than automation alone. Phase 2 analysed 26 hips after excluding candidates with unusable localisation priors, correlating each feature with torsion within diagnosis. Of 213 coecients one survived false-discovery correction, femoral mid-axis extent in the seven expert-annotated HD hips (r=−0.96,q=0.012), and it did not replicate atn=26; the labral descriptors held instead (HD widthr=0.83then0.66). Every principal basis was tted per cohort, so only the named descriptors are compared between cohorts. The result is a governed anatomical modelling workow for digital-twin research, not a dynamic digital twin.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/DT4H_039.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/DT4H_039_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=Ybxz2aM8sU
BibTex
@InProceedings{GarJoh_From_MICCAISAT2026,
author = { Garcia-Henao, John AND Juchler, Norman AND Guerrero-Toro, Cindy AND Bünger, Nicholas AND Herzog, Benedikt AND Vella, Benjamin AND Kunz, Dominik AND Hoch, Armando AND Hirsch, Sven AND Caprara, Sebastiano},
title = { { From PACS to Morphometrics: A Hospital-Integrated Workflow for Hip Digital Twin Research } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17275},
month = {pending},
page = {pending}
}