Abstract
Multicenter cardiac magnetic resonance (CMR) cohorts frequently exhibit pathology-dependent supervision: scar labels may be available for subjects whose edema labels or edema-sensitive sequences are absent. Treating this asymmetry as ordinary missing data either discards reliable supervision or converts an unobserved pathology into false background. We propose a reliability-aware framework that decouples supervision and probabilistically recouples context. Scar and edema are learned through target-specific data streams gated by annotation availability. A first-stage model factorizes its prediction into soft myocardial support and target-lesion confidence, while a full-field refinement model reconnects these fields with the original CMR so that prior errors remain correctable. For edema, scar probability and a myocardium-gated T2 hyperintensity field provide cross-pathology and modality-specific context without imposing a deterministic spatial relationship. Anatomy-constrained residual transplantation further varies lesion morphology and local contrast while preserving image–label consistency. The framework uses nnU-Net as a common backbone, isolating the effects of supervision reliability and probabilistic context from architectural customization. On the official validation leader board, the final system achieved Dice/HD values of 0.73/14.78 mm for scar and 0.72/21.54 mm for edema. The code is available at https://github.com/jingkunchen/MICCAI_CARE_2026
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CARE_047.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=vx0CWICnKg
BibTex
@InProceedings{WanChe_Decoupling_MICCAISAT2026,
author = { Wang, Chenxu AND Zhu, Rongxin AND Chen, Xiaoan AND Luo, Shengda AND Chen, Jingkun},
title = { { Decoupling Supervision and Recoupling Context: Reliability-Aware Soft-Prior Learning for Myocardial Scar and Edema 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}
}