Abstract
We present a compact inference system for the CMR-MULTI cardiac magnetic resonance (CMR) challenge. The task combines multi-view segmentation with two derived clinical quantities: ejection fraction (EF) from cine short-axis images and scar mass from late gadolinium enhancement (LGE) short-axis images. The system treats these outputs as acquisition-specific problems rather than one shared segmentation task. It uses 2D nnU-Net v2 ensembles for cine and long-axis anatomy, a 3D LGE short-axis scar branch, conservative scar refinement, and deterministic EF and scar-mass calculation. On the official validation server, the submitted validation configuration reached an Overall score of 0.712815, with 0.687630 for Task 1 and 0.738000 for Task 2. The Docker image produced the complete 82-file output tree for a 15-case local run in 378 seconds with network access disabled. The result is a simple, auditable path from CMR images to challenge-ready masks and measurements.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRSeg_018.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=9Iv8WzvCLu
BibTex
@InProceedings{CheYix_AcquisitionStructured_MICCAISAT2026,
author = { Chen, Yixin AND Peng, Yizhen},
title = { { Acquisition-Structured nnU-Net Ensembles for Multi-Sequence Cardiac MRI Segmentation and Quantification } },
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}
}