Abstract
The CMR-MULTI challenge requires joint cine segmenta- tion (Task 1) across the short-axis (SAX), two-chamber (2CH) and four- chamber (4CH) views, and late gadolinium enhancement (LGE, Task 2) across the SAX, 2CH, 4CH and right atrium short-axis (RAS) views, together with clinical readings of ejection fraction (EF) and myocardial scar mass. We present a view-specialized hybrid pipeline that assigns the strongest available model to each anatomical perspective rather than enforcing a single architecture globally. Cine SAX is segmented with nnU-Net [4]; cine 2CH/4CH with CineMA fine-tuned [1]; LGE struc- ture with nnU-Net; and LGE SAX scar with a dedicated ScarElastic network [8] operating on nnU-Net myocardium at probability threshold 0.85, followed by largest-component filtering. On non-SAX LGE views we further drop scar connected components smaller than 250 voxels. Clinical EF is recomputed from cine SAX masks; scar mass is derived from LGE SAX with affine calibration max(0, 0.5 m − 4.0). On the challenge vali- dation metric, our submitted hybrid achieves an overall score of 0.7342 (Task 1: 0.7030, Task 2: 0.7655), with cine Dice similarity coefficient (DSC) 0.899, EF Pearson correlation coefficient (PCC) 0.975, LGE DSC 0.767, and scar-mass volume PCC (vPCC) 0.927. We describe training protocols, threshold selection, and development ablations.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRSeg_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=4lqLRaUTPV
BibTex
@InProceedings{ZenWei_ViewSpecialized_MICCAISAT2026,
author = { Zeng, Weichu AND Zheng, Dashun AND Zou, Mingye AND Meng, Mingyuan},
title = { { View-Specialized Hybrid Segmentation for Multi-View Cardiac MRI: Combining nnU-Net, CineMA, and Scar-Elastic Refinement } },
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}
}