Abstract
Multi-sequence cardiac magnetic resonance (CMR) provides complementary tissue contrasts. However, myocardial edema and scar remain difficult to segment because they are small, poorly delineated, and may be discontinuous across slices. Cross-center intensity shifts and imperfect inter-sequence alignment further complicate the joint use of late gadolinium enhancement (LGE), T2-weighted, and balanced steady-state free precession (bSSFP) images. Query-based segmenters can localize irregular lesions, but slice-wise query predictions do not explicitly enforce volumetric coherence. To address this limitation, we propose MaskSAM-PBPR, a two-stage framework that separates query-based localization from prior-guided volumetric refinement. In Stage 1, the Enhanced MaskSAM Prior Generator uses MedSAM initialization, prompt denoising, and Fourier-domain augmentation to produce class-specific soft priors. In Stage 2, the Prior-Guided Bidirectional Propagation Refiner (PBPR) combines Anchor-Guided Class-Wise Propagation with 3D Residual Refinement. This stage converts slice-wise soft priors into volumetric predictions by propagating class-specific evidence across slices and correcting residual local errors at full resolution. On the CARE 2026 validation set, MaskSAM-PBPR achieved scar and edema Dice scores of 0.7200 and 0.7322, with Hausdorff distance (HD) values of 12.5062 and 19.7994 mm.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CARE_014.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=XpVjbaycj6
BibTex
@InProceedings{ZhaQin_MaskSAMPBPR_MICCAISAT2026,
author = { Zhang, Qinghe AND Peng, Yanjun AND Zhang, Xiaoning AND Qiu, Fuzhuang},
title = { { MaskSAM-PBPR: A Prior-Guided Bidirectional Propagation Refiner for Multi-Sequence CMR Myocardial Pathology 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}
}