CoChat
Sign Up
Cited by 20ECG Monitoring and AnalysisMachine Learning in HealthcareArtificial Intelligence in Healthcare and Education

Foundation models for electrocardiogram interpretation: clinical implications

Open access
Nolin-Lapalme, Alexis, Sowa, Achille, Delfrate, Jacques, Tastet, Olivier, Corbin, Denis, Kulbay, Merve, Ozdemir, Derman, Noël, Marie-Jeanne +25 more·2026·European Heart Journal
DOI 10.1093/eurheartj/ehaf1119PMC PMC13178678
Open this paper in CoChat — free account

Abstract

Background and Aims: The 12-lead electrocardiogram (ECG) remains a cornerstone of cardiac diagnostics, yet existing artificial intelligence (AI) solutions for automated interpretation often lack generalizability, remain closed source, and are primarily trained using supervised learning (SL), which requires extensive labelled datasets and may limit adaptability across diverse clinical settings. Self-supervised learning (SSL) can potentially overcome these limitations by learning robust representations from unlabelled data. To address these challenges, this study developed and compared two open-source foundational ECG models: DeepECG-SL, a supervised multilabel ECG model, and DeepECG-SSL, a self-supervised model. Methods: Both models were trained on over 1 million ECGs using a standardized preprocessing pipeline and automated free-text extraction from ECG reports to predict 77 cardiac conditions. DeepECG-SSL leveraged unlabelled data through self-supervised contrastive learning and masked lead modelling before fine-tuning for downstream tasks, while DeepECG-SL was trained directly on labelled diagnostic data in an end-to-end fashion. Performance was evaluated across seven private, multilingual healthcare systems and four public ECG repositories, with assessment of fairness by age and sex, and investigation of privacy vulnerabilities as well as memory and compute requirements. Results: DeepECG-SSL achieved micro-averaged area under the receiver operating characteristic curves (AUROCs) across all 77 cardiac conditions for ECG interpretation of 0.990 [95% confidence interval (CI): 0.990, 0.990] on the internal dataset (MHI-ds), 0.981 (95% CI: 0.981, 0.981) on external public datasets (UKB, CLSA, MIMIC-IV and PTB), and 0.983 (95% CI: 0.983, 0.983) on external private datasets (UW, UCSF, JGH, NYP, MGH, CSH and CHUM), while DeepECG-SL demonstrated AUROCs of 0.992 (95% CI: 0.992, 0.992), 0.980 (95% CI: 0.980, 0.980), and 0.983 (95% CI: 0.983, 0.984), respectively. Fairness analyses revealed minimal disparities (true-positive rate and false-positive rate difference <0.1) across age and sex groups for both models. DeepECG-SSL demonstrated superior performance on limited-data digital biomarker tasks, with the largest improvements in long QT syndrome (LQTS) genotype classification (AUROC 0.931 vs 0.850, P = .026, n = 127 ECGs) and 5 year atrial fibrillation risk prediction (AUROC 0.742 vs 0.734, P < 0.001, n = 132 050 ECGs), while achieving superior performance in left ventricular ejection fraction ≤40% classification (AUROC 0.926 vs 0.917, P < 0.001, n = 25 252 ECGs) and comparable performance in LQTS detection (AUROC 0.767 vs 0.735, P = 0.117, n = 934 ECGs). Conclusions: This study establishes SSL as a promising paradigm for ECG analysis, particularly in settings with limited annotated data, enhancing accessibility, generalizability, and fairness in AI-driven cardiac diagnostics. By releasing model weights, preprocessing tools, and validation code, this work aims to support robust, data-efficient AI diagnostics across diverse clinical environments and questions.

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

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

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

Integrity signals

Registry

Crossref · 10.1093/eurheartj/ehaf1119

Cites

5 of 5 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

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

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

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?

  • ECG foundation models may improve transportability and label efficiency, particularly for lower-prevalence phenotypes, but their translational value will depend on whether they improve calibration, robustness, and pathway performance rather than benchmark metrics alone.

    “ECG foundation models may improve transportability and label efficiency, particularly for lower-prevalence phenotypes, but their translational value will depend on whether they improve calibration, robustness, and pathway performance rather than benchmark metrics alone.68”

    From AI-ECG for Echocardiography Triage in Structural Heart Disease: Evidence, Implementation, and Future Directions (2026) · cites this paper as [68]

Self-check

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

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

Cites

Flagged first, then by citations

Research works printed but not linked: 43

  1. [1]Holst H, Ohlsson M, Peterson C, Edenbrandt L. A confident decision support system for interpreting electrocardiograms. Clin Physiol 1999; 19: 410– 8. 10.1046/j.1365-2281.1999.00195.x 10516892 · DOI
  2. [2]Martínez-Sellés M, Marina-Breysse M. Current and future use of artificial intelligence in electrocardiography. J Cardiovasc Dev Dis 2023; 10: 175. 10.3390/jcdd10040175 37103054 PMC10145690 · DOI
  3. [3]Hannun AY, Rajpurkar P, Haghpanahi M, Tison GH, Bourn C, Turakhia MP, et al Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nat Med 2019; 25: 65– 9. 10.1038/s41591-018-0268-3 30617320 PMC6784839 · DOI · Paper page
  4. [4]Yao X, Rushlow DR, Inselman JW, McCoy RG, Thacher TD, Behnken EM, et al Artificial intelligence–enabled electrocardiograms for identification of patients with low ejection fraction: a pragmatic, randomized clinical trial. Nat Med 2021; 27: 815– 9. 10.1038/s41591-021-01335-4 33958795 · DOI · Paper page
  5. [5]Aras MA, Abreau S, Mills H, Radhakrishnan L, Klein L, Mantri N, et al Electrocardiogram detection of pulmonary hypertension using deep learning. J Card Fail 2023; 29: 1017– 28. 10.1016/j.cardfail.2022.12.016 36706977 PMC10363571 · DOI · Paper page
  6. [6]Jiang R, Cheung CC, Garcia-Montero M, Davies B, Cao J, Redfearn D, et al Deep learning–augmented ECG analysis for screening and genotype prediction of congenital long QT syndrome. JAMA Cardiol 2024; 9: 377– 84. 10.1001/jamacardio.2024.0039 38446445 PMC10918571 · DOI
  7. [7]LeCun Y, Bengio Y, Hinton G. Deep learning. Nature 2015; 521: 436– 44. 10.1038/nature14539 26017442 · DOI · Paper page
  8. [8]Krishnan R, Rajpurkar P, Topol EJ. Self-supervised learning in medicine and healthcare. Nat Biomed Eng 2022; 6: 1346– 52. 10.1038/s41551-022-00914-1 35953649 · DOI · Paper page
  9. [9]Lai J, Tan H, Wang J, Ji L, Guo J, Han B, et al Practical intelligent diagnostic algorithm for wearable 12-lead ECG via self-supervised learning on large-scale dataset. Nat Commun 2023; 14: 3741. 10.1038/s41467-023-39472-8 37353501 PMC10290151 · DOI
  10. [10]Mehari T, Strodthoff N. Self-supervised representation learning from 12-lead ECG data. Comput Biol Med 2022; 141: 105114. 10.1016/j.compbiomed.2021.105114 34973584 · DOI
  11. [12]Kelly BS, Judge C, Bollard SM, Clifford SM, Healy GM, Aziz A, et al Radiology artificial intelligence: a systematic review and evaluation of methods (RAISE). Eur Radiol 2022; 32: 7998– 8007. 10.1007/s00330-022-08784-6 35420305 PMC9668941 · DOI · Paper page
  12. [13]Pineau J, Vincent-Lamarre P, Sinha K, Larivière V, Beygelzimer A, d’Alché-Buc F, et al Improving reproducibility in machine learning research (A Report from the NeurIPS 2019 Reproducibility Program). arXiv, 10.48550/arXiv.2003.12206, 30 December 2020, preprint: not peer reviewed. · DOI · Paper page
  13. [14]Oh J, Chung H, Kwon J, Hong D, Choi E. Lead-agnostic self-supervised learning for local and global representations of electrocardiogram. arXiv, 10.48550/arXiv.2203.06889, 18 March 2022, preprint: not peer reviewed. · DOI
  14. [15]World Medical Association. World medical association declaration of Helsinki: ethical principles for medical research involving human subjects. JAMA 2013; 310: 2191– 4. 10.1001/jama.2013.281053 24141714 · DOI · Paper page
  15. [18]Raina P, Wolfson C, Kirkland S, Griffith LE, Balion C, Cossette B, et al Cohort profile: the Canadian Longitudinal Study on Aging (CLSA). Int J Epidemiol 2019; 48: 1752– 3j. 10.1093/ije/dyz173 31633757 PMC6929533 · DOI
  16. [19]Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, et al UK biobank: an open access resource for identifying the causes of a wide range of Complex diseases of middle and old age. PLoS Med 2015; 12: e1001779. 10.1371/journal.pmed.1001779 25826379 PMC4380465 · DOI · Paper page
  17. [20]Wagner P, Strodthoff N, Bousseljot R-D, Kreiseler D, Lunze FI, Samek W, et al PTB-XL, a large publicly available electrocardiography dataset. Sci Data 2020; 7: 154. 10.1038/s41597-020-0495-6 32451379 PMC7248071 · DOI
  18. [21]Gajšek P, Ravazzani P, Grellier J, Samaras T, Bakos J, Thuróczy G. Review of studies concerning Electromagnetic Field (EMF) exposure assessment in Europe: low frequency fields (50 Hz–100 kHz). Int J Environ Res Public Health 2016; 13: 875. 10.3390/ijerph13090875 27598182 PMC5036708 · DOI
  19. [22]Chavan MS, Agarwala RA, Uplane MD. Suppression of noise in the ECG signal using digital IIR filter. In: Proceedings of the 8th WSEAS International Conference on Multimedia Systems and Signal Processing, p. 335–43. Stevens Point, Wisconsin, USA: World Scientific and Engineering Academy and Society (WSEAS); 2008.
  20. [23]Greutmann M, Horlick EM. Uncommon cause for a common electrocardiographic artifact. Can J Cardiol 2010; 26: e30. 10.1016/s0828-282x(10)70344-5 20101367 PMC2827235 · DOI
  21. [24]Jacoby WG. Loess. Elect Stud 2000; 19: 577– 613. 10.1016/S0261-3794(99)00028-1 · DOI
  22. [25]Kligfield P, Gettes LS, Bailey JJ, Childers R, Deal BJ, Hancock EW, et al Recommendations for the standardization and interpretation of the electrocardiogram: part I: the electrocardiogram and its technology a scientific statement from the American Heart Association Electrocardiography and Arrhythmias Committee, Council on Clinical Cardiology; the American College of Cardiology Foundation; and the Heart Rhythm Society endorsed by the International Society for Computerized Electrocardiology. J Am Coll Cardiol 2007; 49: 1109– 27. 10.1016/j.jacc.2007.01.024 17349896 · DOI
  23. [27]Devlin J, Chang M-W, Lee K, Toutanova K. BERT: pre-training of deep bidirectional transformers for language understanding. arXiv, 10.48550/arXiv.1810.04805, 24 May 2019, preprint: not peer reviewed. · DOI · Paper page
  24. [28]Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN, et al Attention is all you need. arXiv, 10.48550/arXiv.1706.03762, 2 August 2023, preprint: not peer reviewed. · DOI · Paper page
  25. [31]Baevski A, Zhou H, Mohamed A, Auli M. wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations. arXiv, 10.48550/arXiv.2006.11477, 22 October 2020, preprint: not peer reviewed. · DOI · Paper page
  26. [32]Kiyasseh D, Zhu T, Clifton DA. CLOCS: contrastive learning of cardiac signals across space, time, and patients. arXiv, 10.48550/arXiv.2005.13249, 16 May 2021, preprint: not peer reviewed. · DOI
  27. [33]Jabbour G, Nolin-Lapalme A, Tastet O, Corbin D, Jordà P, Sowa A, et al Prediction of incident atrial fibrillation using deep learning, clinical models, and polygenic scores. Eur Heart J 2024; 45: 4920– 34. 10.1093/eurheartj/ehae595 39217446 PMC11631091 · DOI
  28. [34]Hardt M, Price E, Srebro N. Equality of opportunity in supervised learning. arXiv, 10.48550/arXiv.1610.02413, 7 October 2016, preprint: not peer reviewed. · DOI · Paper page
  29. [35]Li J, Aguirre A, Moura J, Liu C, Zhong L, Sun C, et al An electrocardiogram foundation model built on over 10 million recordings. NEJM AI 2025; 2. https://ai.nejm.org/doi/abs/10.1056/AIoa2401033 7 PMC12327759 40771651 · DOI
  30. [36]McKeen K, Oliva L, Masood S, Toma A, Rubin B, Wang B. ECG-FM: an open electrocardiogram foundation model. arXiv, 10.48550/arXiv.2408.05178, August 2024, preprint: not peer reviewed. PMC12530324 41113504 · DOI
  31. [38]van der Maaten L, Hinton G. Visualizing data using t-SNE. J Mach Learn Res 2008; 9: 2579– 605. https://api.semanticscholar.org/CorpusID:5855042.
  32. [39]Ribeiro MT, Singh S, Guestrin C. ‘Why should i trust you?’: explaining the predictions of any classifier. 10.1145/2939672.2939778. 2016. · DOI · Paper page
  33. [40]Selvaraju RR, Cogswell M, Das A, Vedantam R, Parikh D, Batra D. Grad-CAM: visual explanations from deep networks via gradient-based localization. Int J Comput Vis 2020; 128: 336– 59. 10.1007/s11263-019-01228-7 · DOI · Paper page
  34. [41]DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988; 44: 837– 45. 10.2307/2531595 3203132 · DOI · Paper page
  35. [42]Zhu K, Zheng Y, Chan KCG. Weighted brier score—an overall summary measure for risk prediction models with clinical utility consideration. Stat Biosci 2025: 10.1007/s12561-025-09505-5. 10.1007/s12561-025-09505-5. PMC12523994 41103457 · DOI
  36. [43]Pencina MJ, D’ Agostino RB Sr, D’ Agostino RB Jr, Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med 2008; 27: 157– 72. 10.1002/sim.2929 17569110 · DOI · Paper page
  37. [44]Tan M, Le QV. EfficientNetV2: smaller models and faster training. arXiv, 10.48550/arXiv.2104.00298, 23 June 2021, preprint: not peer reviewed. · DOI
  38. [45]Adebayo J, Gilmer J, Muelly M, Goodfellow I, Hardt M, Kim B. Sanity checks for saliency maps. arXiv, 10.48550/arXiv.1810.03292, 5 November 2020, preprint: not peer reviewed. · DOI · Paper page
  39. [46]Tian Y, Li Z, Jin Y, Wang M, Wei X, Zhao L, et al Foundation model of ECG diagnosis: diagnostics and explanations of any form and rhythm on ECG. Cell Rep Med 2024; 5: 101875. 10.1016/j.xcrm.2024.101875 39694017 PMC11722092 · DOI
  40. [48]Hou Y, Fan Z, Li J, Zeng Z, Lv G, Lin J, et al Deep learning-based 12-lead electrocardiogram for low left ventricular ejection fraction detection in patients. Can J Cardiol 2024; 41: 278– 90. 10.1016/j.cjca.2024.09.018. 39343388 · DOI
  41. [49]Himmelreich JCL, Lucassen WAM, Harskamp RE, Aussems C, Van Weert HCPM, Nielen MMJ. CHARGE-AF in a national routine primary care electronic health records database in the Netherlands: validation for 5-year risk of atrial fibrillation and implications for patient selection in atrial fibrillation screening. Open Heart 2021; 8: e001459. 10.1136/openhrt-2020-001459 33462107 PMC7816907 · DOI
  42. [51]Suh J, Kim J, Jung E, Rhee W. Evaluating feature attribution methods for electrocardiogram. arXiv, 10.48550/arXiv.2211.12702, 22 October 2024, preprint: not peer reviewed. · DOI
  43. [56]Dwork C. Differential privacy. In: Bugliesi M, Preneel B, Sassone V, Wegener I (eds.), Automata, Languages and Programming. Berlin, Heidelberg: Springer Berlin Heidelberg, 2006, 1– 12.

Cited by

Sorted by citations

Authors
Nolin-Lapalme, A., Sowa, A., Delfrate, J., Tastet, O., Corbin, D., Kulbay, M., Ozdemir, D., Noël, M.-J., Marois-Blanchet, F.-C., Harvey, F., Sharma, S., Ansari, M., Chiu, I. M., D’souza, V., Friedman, S. F., Chassé, M., Potter, B. J., Afilalo, J., Elias, P. A., Jabbour, G., Bahani, M., Dubé, M.-P., Boyle, P. M., Chatterjee, N. A., Barrios, J., Tison, G. H., Ouyang, D., Maddah, M., Khurshid, S., Cadrin-Tourigny, J., Tadros, R., Hussin, J., & Avram, R.
Affiliations
  • Montreal Heart Institute (CA)
  • Centre Hospitalier de l’Université de Montréal (CA)
  • University of Washington (US)
  • University of California, San Francisco (US)
  • Cedars-Sinai Medical Center (US)
  • Data Sciences Platform, The Broad Institute of MIT and Harvard , Cambridge, MA ,
  • Broad Institute (US)
  • Jewish General Hospital (CA)
  • Columbia University Irving Medical Center (US)
  • Montreal Heart Institute , 5000 Rue Bélanger, Montreal, Quebec , H1T 1C8
Corresponding author
Avram, Robert
ORCID
Venue
European Heart Journal
Publisher
Oxford University Press (OUP)
Published
2026-01-22
Volume
47
Issue
18
Pages
2174-2186
ISSN
  • 0195-668X (ISSN-L)
  • 1522-9645
DOI
10.1093/eurheartj/ehaf1119
PMC
PMC13178678
OpenAlex
W7125391335
Cited by
20
References
5
Citations by year
2026: 19 · 2025: 1
FWCI
13.89
Type
journal article
Language
en
Open access
Licence
Locations
  • European Heart Journal (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
Funders
  • Fonds de la recherche du Québec en Santé (312758)
  • Des Groseillers-Bérard Interventional Cardiology Research
  • Chair
  • Canadian Institute for Advanced Research
  • IVADO Data Science Institute
  • Fonds de recherche du Québec
  • Nature et technologies
  • Canadian Institutes of Health Research
  • Canadian Research Chair Tier 2 in Responsible Multi-Omics Data Science
  • NIH (K23HL169839, 3OT2OD035404-01S3, DP2HL174046)
  • AHA (23CDA1050571, 961045)
Topics
  • Health Sciences › Medicine › Cardiology and Cardiovascular Medicine › ECG Monitoring and Analysis
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
Keywords
ECG interpretation, self-supervised learning, contrastive learning, model fairness, privacy vulnerabilities, model generalization, clinical implementation, multi-label classification, clinical decision support
MeSH terms
Supervised Machine Learning, Electrocardiography, Humans

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