Abstract
Multicentre cardiac MR datasets may independently lack imaging sequences and annotations. We propose APL-Base, which masks losses for unobserved targets and predicts nested scar, pathology parent, and myocardial wall targets. Nested anatomical projection (NAP) restricts pathology predictions to myocardial support, while Component-based Scar Rescue (CSR) converts predictions from an LGE-only expert into residual scar proposals within the fixed parent. On 15 official validation cases, the full pipeline achieved scar and edema Dice scores of 0.6722 and 0.7109. In a five-fold, patient-level out-of-fold (OOF) evaluation of 220 patients, NAP reduced scar HD95 by 11.22 mm but decreased sensitivity, whereas rescue increased scar Dice by approximately 0.02. Among the 80 patients with both pathology labels, this scar gain was offset by lower edema overlap and did not improve equally weighted joint Dice. These results show that anatomical projection and LGE proposal rescue provide complementary boundary control and scar recovery. They also indicate that parent and proposal coverage, rather than selector complexity, limit further improvement.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CARE_005.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=gJ0V3jCW8L
BibTex
@InProceedings{GaoXin_Myocardial_MICCAISAT2026,
author = { Gao, Xinyi AND Zhao, Yuming AND Crean, Andrew AND Costen, Fumie},
title = { { Myocardial Scar and Edema Segmentation with Partial Labels and LGE-Guided Rescue } },
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}
}
