Abstract
Automated cardiac magnetic resonance (CMR) analysis must generalise across imaging sequences (cine and late gadolinium enhancement, LGE), acquisition views (short-axis, two- and four-chamber), and clinical tasks (segmentation and biomarker quantification). We describe a reproducible pipeline for the CMR-MULTI challenge (MWM4MICCAI 2026) built on view-specialised 2D nnU-Net models. Our central observation is that the released cine volumes interleave slices and cardiac phases on a single axis, so a spacing-based planner selects the wrong slice axis; a geometry-aware reformatting rewrites the voxel spacing to restore anatomically correct 2D training. We further derive left-ventricular ejection fraction (LVEF) and LGE scar mass directly from the predicted SAX segmentations using the acquisition metadata, and show that both estimators reproduce the reference biomarkers exactly when applied to ground-truth masks, validating the measurement pipeline. On the public validation split a five-fold ensemble attains an overall score of 0.68 (cine Dice 0.91, LGE Dice 0.75, ejection-fraction Pearson 0.89; scar-mass validation is thin, n=7 with five non-zero-scar cases, so Task-2 numbers are indicative), up from 0.66 for a single fold as the boundary-sensitive Hausdorff term falls. We analyse the scoring behaviour (the maximum-Hausdorff term is strongly saturated; scar is the hardest structure) and report ablations with patient-clustered bootstrap intervals.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRSeg_042.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/CMRSeg_042_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=8fgjDHh6FH
BibTex
@InProceedings{AntEka_An_MICCAISAT2026,
author = { Antipushina, Ekaterina AND Sukhorukov, Daniil AND Cherkasskaya, Elizaveta AND Kalimullin, Ruslan AND Makarov, Ilya AND Koush, Yury},
title = { { An Auditable Geometry Contract for Multi-View Cardiac Magnetic Resonance Imaging Analysis } },
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}
}