Abstract
Automated multi-view cardiac MRI segmentation and clinical biomarker estimation (LVEF and myocardial mass) are limited by sequence heterogeneity (Cine and LGE), extreme class imbalance (pathological scar), and gradient conflicts in monolithic multi-head architectures. We propose an anatomy-prior guided, decoupled single-task expert network system. To eliminate gradient conflicts, we decouple the multi-view baseline into independent expert models optimized via 5-fold cross-validation. To enforce anatomical priors, we deploy a 3D Swin UNETR on native-resolution spatiotemporal Cine stacks (Task 1) to prevent temporal-spatial collapse, and utilize a 2D TransUNet with directional spatial augmentations (disabling vertical flips on long-axis views to preserve top-bottom anatomical prior) and slice-level normalization for LGE (Task 2). On the official MICCAI 2026 CMR-MULTI validation set, our 3D Cine ensemble achieves Dice scores of 85.31% (SAX), 82.39% (2CH), and 84.68% (4CH), with HD95 errors tightly constrained to 2.36 mm. Our 2D LGE ensemble achieves Dice scores of 66.32% (SAX), 63.31% (2CH), 76.56% (4CH), and 83.71% (RAS). Containerized Docker deployment on the unlabeled test set yields physiologically compliant clinical estimations (LVEF: 27.0%-59.8%; Myocardial Mass: 73.9g-170.9g), demonstrating robust generalizability and clinical utility.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRSeg_076.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/CMRSeg_076_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=KN7kao094b
BibTex
@InProceedings{LiuWen_AnatomyPrior_MICCAISAT2026,
author = { Liu, Wenpei AND Zhang, Zhihan AND Yang, Zehao AND Song, Yujia AND Li, Chenxi AND Liu, Xinyang AND Sun, Yulin AND He, Runnan AND Liu, Xiuyun AND Ming, Dong},
title = { { Anatomy-Prior Guided Decoupled Expert Networks for Multi-View Cardiac MRI Segmentation and Biomarker Estimation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17268},
month = {pending},
page = {pending}
}