Abstract
Coronary angiography (CAG) provides detailed anatomical information on the location and severity of epicardial coronary stenosis and remains central to risk stratification and treatment planning in coronary artery disease. However, CAG is invasive and unsuitable as a first-line test for all patients with suspected coronary disease. Magnetocardiography (MCG) is a contactless and non-invasive technique that measures cardiac electromagnetic activity and may capture subtle electrophysiological abnormalities associated with coronary stenosis. This motivates the development of non-invasive models that use MCG to predict CAG-defined coronary stenosis. We propose a multi-label framework motivated by two assumptions: regional signal structure matters for vessel localization, and vessel predictions should remain consistent with global CAG positivity. We design a transformer-based student network to explore regional relationships among sensors and cardiac temporal dynamics. A frozen ResNet1D teacher provides stable hierarchy-projected soft targets for BCE-based distillation. At inference time, validation-set threshold calibration is combined with the same clinical hierarchy to obtain final binary decisions, enforcing consistency between global and vessel-level decisions. On the external test cohort with 362 MCG recordings, the proposed framework achieved AUCs of 0.97 for CAG prediction and 0.85 on average for vessel-level predictions. These results suggest a strong probability ranking, while hierarchy-guided calibration prevents vessel-positive/CAG-negative binary decisions. 2 R. Zhang et al.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_032.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=K2MOtYdy2N
BibTex
@InProceedings{ZhaRui_Clinical_MICCAISAT2026,
author = { Zhang, Rui AND Jia, Dengqiang AND Zhang, Shulin AND Wang, Yuexia AND Kang, Mengyang AND Zhou, Rushuang AND Dong, Yining AND Shen, Dinggang},
title = { { Clinical Hierarchy and Regional Awareness for Multi-Label Classification of Magnetocardiograms } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17272},
month = {pending},
page = {pending}
}