Abstract
Accurate cardiac magnetic resonance imaging (CMR) segmentation is essential for quantifying clinical biomarkers such as ejection fraction, myocardial mass, and scar burden. However, universal CMR segmentation remains challenging due to variations across pulse sequences, anatomical views, scanners, institutions, and label configurations. To address these challenges, we propose NATA-Net, a lightweight multi-head neural network for multi-sequence and multi-view CMR segmentation. The model uses a shared fully convolutional dense encoder–decoder backbone to learn common cardiac representations, while task-specific prediction heads adapt the output space to different modalities and views. This unified design avoids training separate models for each task and allows all available datasets to be jointly exploited, which is beneficial under limited annotation settings. With only 2.6M parameters and 67 layers, NATA-Net provides an efficient balance between segmentation accuracy, memory usage, and inference speed.
On the Universal Multi-Sequence, Multi-Center and Multi-View CMR Segmentation Challenge validation set, NATA-Net achieved an average DSC of $0.91$, HD of $8.67$ mm, and ASD of $0.47$ mm for Cine segmentation. For LGE segmentation, it achieved an average DSC of $0.75$ across SAX, 2CH, 4CH, and RAS views. For clinical metric estimation, NATA-Net obtained a PCC of $0.89$ for Cine ejection fraction prediction and a volumetric PCC of $0.70$ for LGE scar mass estimation. Our method achieved an average challenge score of $0.69$.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRSeg_009.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=Ohtif6XF9n
BibTex
@InProceedings{TriMin_NATANet_MICCAISAT2026,
author = { Trinh, Minh Nhat AND Correia, Teresa M.},
title = { { NATA-Net: A Network with Adaptive Task-Aware Heads for Universal 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}
}