Abstract
Accurate segmentation of cardiac structures and myocardial scar from cardiovascular magnetic resonance (CMR) imaging is essential for quantitative as-sessment of cardiac function and tissue characterization. However, develop-ing a unified model that generalizes across multiple imaging sequences, anatomical views, and acquisition protocols remains challenging due to substantial variations in image appearance, spatial resolution, and class imbalance. In this work, we present our solution for the Universal Multi-Sequence, Multi-Center, and Multi-View Cardiac MRI Challenge based on an enhanced xLSTM-UNet framework. The proposed architecture combines hierarchical convolutional feature extraction with UxLSTM bottleneck modules to jointly capture local anatomical details and long-range spatial dependencies. To improve segmentation performance, we introduce a cardiac-specific hybrid loss that integrates Dice, Cross-Entropy, TopK Cross-Entropy, and Boundary losses, enabling accurate delineation of thin myocardial structures and small scar regions. Furthermore, we design an MRI-oriented augmentation strategy incorporating elastic deformation, Rician noise, bias-field simulation, cardiac motion blur, and intensity transformations to improve robustness across heterogeneous cine and late gadolinium enhancement (LGE) MRI acquisitions. Separate 2D and 3D models were trained for cine and LGE MRI according to their imaging characteristics within a unified framework. The proposed approach demonstrated strong performance on the challenge validation leaderboard, achieving 2nd place overall, highlighting its robustness and generalization across multiple cardiac sequences, imaging views, and clinical evaluation tasks.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRSeg_026.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=QJ1dPVAo34
BibTex
@InProceedings{QayAbd_CardioMix_MICCAISAT2026,
author = { Qayyum, Abdul AND Mazher, Moona AND Niederer, Steven A.},
title = { { CardioMix: A Unified Framework for Multi-Sequence and Multi-View Cardiac MRI Segmentation } },
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}
}