Abstract
Robust multi-chamber segmentation is essential for reliable cardiac function assessment. A clinically useful pipeline must accurately delineate cardiac chambers while preserving two structural constraints: intra-view anatomical topology and cross-view consistency. We propose CATCH, a segmentation framework that implicitly models both constraints by combining self-supervised pretrained representations with a cross-view joint self-attention module. This module captures the anatomical relationships between the apical two-chamber (A2C) and apical four-chamber (A4C) views. Because supervision is required for only one view per pair, the framework learns effectively from partially annotated data. On an internal benchmark, CATCH performs on par with state-of-the-art methods while achieving the best left-ventricle Dice (91.49%) and HD95 (5.11 mm). In a zero-shot evaluation on CAMUS, it generalizes better than MedSAM, reaching 92.97% LV Dice and an ejection-fraction MAE of 3.72. Qualitative results further demonstrate more anatomically and topologically consistent masks. The code is available at https://github.com/DeepRCL/CATCH.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ASMUS_051.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=3dE2CB3l2e
BibTex
@InProceedings{MedBas_CATCH_MICCAISAT2026,
author = { Medhat, Bassant AND Hashemi, Nima AND Abdelsamad, Baraa AND Chen, Edward S. AND Sojoudi, Samira AND Luong, Christina L. AND Tsang, Teresa S. M. AND Abolmaesumi, Purang},
title = { { CATCH: Cross-View Attention for Consistent Multi-Chamber Heart Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17276},
month = {pending},
page = {pending}
}