Abstract
Multi-center MR whole-heart segmentation is challenging because cardiac chambers, myocardium, and great vessels have different sizes, boundary contrasts, and field-of-view coverage. nnU-Net provides a competitive baseline, but our validation analysis shows that the remaining errors concentrate in myocardium (Myo), aorta (AO), and pulmonary artery (PA). We address these structure-specific challenges with a structure-reliability-guided inference framework. A five-fold nnU-Net prediction is used as the anatomical anchor, AO and Myo are refined by an automatic second-stage ROI branch, and an MAE-pretrained ResEncL nnU-Net with lightweight GraTa-style test-time adaptation serves as a donor for LV, RV, Myo, and PA. The donor assignment is determined from out-of-fold (OOF) validation and fixed before official inference. Donor predictions are accepted only through an add-only rule, while AO, LA, and RA are locked after ROI tuning. On internal five-fold OOF validation, the final pipeline improved mean Dice from 0.8843 to 0.8864. The largest gains were observed on PA (+0.0053), Myo (+0.0038), AO (+0.0027), and RV (+0.0017). On the CARE 2026 official MR validation leaderboard snapshot, the method achieved 0.8668 Dice, 14.3296 HD, and 1.6440 ASSD, ranking first at the recorded submission time. These results suggest that MR whole-heart segmentation can benefit from assigning model responsibility at the anatomical-structure level.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CARE_029.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=6CegVfrwDK
BibTex
@InProceedings{DuYan_StructureReliabilityGuided_MICCAISAT2026,
author = { Du, Yanhong AND Zhang, Xiao AND Chen, Xiaoan AND Feng, Jun AND Chen, Jingkun},
title = { { Structure-Reliability-Guided ROI and Donor Fusion for Multi-Center MR Whole-Heart Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17273},
month = {pending},
page = {pending}
}
