CoChat

Deep learning enhanced magnetocardiography enables multi-task detection of coronary, ventricular, and rhythm disorders

Open access
Kranz, Dominik D., Kahriman, Oruç, Dischl, Dominic, Treskatsch, Sascha, Sander, André, Brachmann, Johannes, Park, Jai-Wun, Wessel, Niels · 2025 · medRxiv

The DOI matches its Crossref record; 33 of the 33 references it cites were found in a registry or an index.

Abstract

Background: Magnetocardiography (MCG) captures the magnetic fields generated by myocardial currents, theoretically preserving electrophysiological details lost in surface potentials. However, its clinical application has been hindered by small datasets and the complexity of analysing high-dimensional magnetic field data. We sought to develop a self-supervised deep learning framework to detect high burden coronary artery disease (CAD), left ventricular dysfunction, and arrhythmia risk from resting MCG recordings. Methods: We developed MCG2Vec, a contrastive deep learning encoder trained on raw 64-channel MCG signals. The model was pre-trained on unlabeled 10-second segments to learn generalizable signal morphology, then fine-tuned for clinical tasks in a retrospective cohort of 1,732 consecutive patients. The primary endpoints were the detection of significant CAD ( ≥ 70% stenosis), reduced left ventricular ejection fraction (LVEF <55%), and atrial fibrillation (AF) risk, all derived from sinus-rhythm recordings. Model performance was evaluated using patient-stratified five-fold cross-validation and interpreted using Grad-CAM activation mapping. Results: In the validation analysis, the model detected significant CAD with an area under the curve (AUC) of 0.89 (95% CI 0.82 - 0.93) and successfully localized ischemia to the left anterior descending (AUC 0.88), right coronary (0.82), and left circumflex arteries (0.82). Reduced LVEF was identified with an AUC of 0.81 (95% CI 0.71 - 0.91). Furthermore, the model predicted a history of paroxysmal AF from sinus-rhythm recordings with an AUC of 0.77 (95% CI 0.69 - 0.85). Explainability analysis confirmed that the model relied on physiologically distinct phases of the cardiac cycle (repolarization heterogeneity for ischemia and P-wave/S-wave morphology for arrhythmia risk) rather than non-specific noise. Conclusion: Deep learning-enhanced magnetocardiography enables the accurate, non-invasive detection of ischemia, ventricular dysfunction, and arrhythmia risk from a single resting scan. By unlocking the latent diagnostic information within the cardiac magnetic field, this approach offers a scalable, radiation-free adjunct to standard electrocardiography for precision cardiac diagnostics.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.64898/2025.11.30.25341301

Cites

33 of 33 references verified0 citation errors · 0 not checkable

Statistics

Figures

Image 1
CC-BY-NC-ND · From the PDF
Figure 2 Overview of the MCG encoder model pretraining and finetuning pipeline. During pretraining, short 14 multichannel MCG segments undergo stochastic augmentations and are encoded by a contrastive encoder 15 (MCG2Vec, a ). The encoder is optimized using a contrastive loss to produce latent embeddings that maximize 16 similarity between augmented views of the same signal while minimizing similarity to embeddings from different 17 signals. 18
Figure 2 Overview of the MCG encoder model pretraining and finetuning pipeline. During pretraining, short 14 multichannel MCG segments undergo stochastic augmentations and are encoded by a contrastive encoder 15 (MCG2Vec, a ). The encoder is optimized using a contrastive loss to produce latent embeddings that maximize 16 similarity between augmented views of the same signal while minimizing similarity to embeddings from different 17 signals. 18CC-BY-NC-ND · From the PDF
Image 3
CC-BY-NC-ND · From the PDF
Figure 4 ( a ) Receiver operating characteristic (ROC) curves showing fine-tuned model performance for CAD, reduced LVEF, and AF classification, all derived from sinus rhythm recordings. Shaded areas indicate 95% confidence intervals. ( b ) Channel-wise averaged beats of the validation sets, color-coded by Grad-CAM intensity, highlighting morphology segments most relevant to model predictions. ( c ) Schematic of coronary arteries (LAD, LCX, RCA) and MCG sensor layout used for spatial localization analysis. ( d ) ROC curves for artery-specific
Figure 4 ( a ) Receiver operating characteristic (ROC) curves showing fine-tuned model performance for CAD, reduced LVEF, and AF classification, all derived from sinus rhythm recordings. Shaded areas indicate 95% confidence intervals. ( b ) Channel-wise averaged beats of the validation sets, color-coded by Grad-CAM intensity, highlighting morphology segments most relevant to model predictions. ( c ) Schematic of coronary arteries (LAD, LCX, RCA) and MCG sensor layout used for spatial localization analysis. ( d ) ROC curves for artery-specificCC-BY-NC-ND · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 33 references

Research works printed but not linked: 2

  1. 10.48550/arXiv.2505.18185 · DOI
  2. C. Yang , M. Westover , and J. Sun , “BIOT: Biosignal Transformer for Cross-data Learning in the Wild,” in Advances in Neural Information Processing Systems, A. Oh , T. Naumann , A. Globerson , K. Saenko , M. Hardt , and S. Levine , Eds., Curran Associates, Inc., 2023, pp. 78240–78260. [Online]. Available: https://proceedings.neurips.cc/paper_files/paper/2023/file/f6b30f3e2dd9cb53bbf2024402d02295-Paper-Conference.pdf

Cited by

See all 1

Metadata

Authors
Kranz, D. D., Kahriman, O., Dischl, D., Treskatsch, S., Sander, A., Brachmann, J., Park, J.-W., & Wessel, N.
Affiliations
  • Humboldt-Universität zu Berlin (DE)
  • Deutsches Herzzentrum München (DE)
  • Charité – Universitätsmedizin Berlin (DE)
  • University of Split (HR)
  • Deutsches Herzzentrum der Charité (DE)
Corresponding author
Wessel, Niels
Venue
medRxiv
Publisher
openRxiv
Published
2025-12-02
OpenAlex
W4417151796
Cited by
1
References
33
Citations by year
2026: 1
Type
posted content
Language
en
Open access
Locations
medRxiv (acceptedVersion, cc-by-nc-nd)
Topics
  • Health Sciences › Medicine › Cardiology and Cardiovascular Medicine › ECG Monitoring and Analysis
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Cardiac Imaging and Diagnostics
Keywords
magnetocardiography, coronary artery disease detection, left ventricular dysfunction, atrial fibrillation risk, ischemia localization, self-supervised contrastive learning, cardiac magnetic field, noninvasive cardiac assessment

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Deep learning enhanced magnetocardiography enables multi-task detection of coronary, ventricular, and rhythm disorders. https://app.cochat.ai/papers/10.64898/2025.11.30.25341301
The paper
Kranz, D. D., Kahriman, O., Dischl, D., Treskatsch, S., Sander, A., Brachmann, J., Park, J.-W., & Wessel, N. (2025). Deep learning enhanced magnetocardiography enables multi-task detection of coronary, ventricular, and rhythm disorders. medRxiv. https://doi.org/10.64898/2025.11.30.25341301

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.