AI-Enabled ECG Found to Detect HCM With High Accuracy

hypertrophic cardiomyopathy
HOCM, hypertrophic cardiomyopathy, cardiomyopathy
An artificial intelligence algorithm trained to assess digital 12-lead electrocardiograms was found to identify hypertrophic cardiomyopathy with high diagnostic accuracy.

An artificial intelligence (AI) algorithm trained to assess digital 12-lead electrocardiograms (ECGs) was found to identify hypertrophic cardiomyopathy (HCM) with high diagnostic accuracy, particularly in younger patients, according to study results published in the Journal of the American College of Cardiology.

Although echocardiography is the main tool used to diagnose HCM, the optimal diagnostic approach in asymptomatic patients remains uncertain. Evaluation of ECG readings by AI-based algorithms using a convolutional neural network (CNN) may allow the detection of features that would otherwise be missed by expert human interpreters.

In this study, 3060 patients with verified HCM (mean age, 54.8±15.9 years; 44.3% women) and 63,941 age- and sex-matched control individuals without HCM (mean age, 57.5±15.5 years; 43.1% women) were enrolled. Participants were divided into algorithm training (n=46,901), validation (n=6700), and testing (n=13,400) groups. Participants were given a standard supine 12-lead digital ECGs between July 1987 and November 2017. The CNN was then applied for ECG interpretation. The study’s primary outcome was the ability of the CNN to detect HCM, solely based on the 12-lead ECG results.

The algorithm training and validation were conducted on the data of 2448 patients with HCM and 51,153 control individuals. The area under the curve (AUC) of the CNN after training and validation was 0.95 in the validation group (95% CI, 0.94-0.97) and 0.96 (95% CI, 0.95-0.96) in the testing group (sensitivity, 87%; specificity, 90%; patients with HCM, n=612; patients without HCM, n=12,788), using 11% as the probability threshold for having HCM.

Related Articles

The model was found to perform best in patients <40 years, with a sensitivity and specificity of 95% and 92%, respectively (diagnostic odds ratio, 195.0; 95% CI, 84.3-451.2). In patients with indications of left ventricular hypertrophy on ECG, the AUC was 0.95 (95% CI, 0.94-0.97), compared with an AUC of 0.95 (95% CI, 0.90-1.00) in those without LVH on ECG. The median probabilities for an HCM diagnosis on ECG in participants with (n=286) and without (n=574) sarcomeric mutations were 97% and 96%, respectively.

Study limitations include a single-center setup, unclear applicability to other settings/populations, a possible underestimation of test accuracy, obscured features “seen” by the CNN, and unavailability of certain patient demographic details.

“This model requires further refinement and external validation, but it may hold promise for HCM screening,” noted the authors.

Funding and Conflicts of Interest Disclosures

The funding source was The Louis V. Gerstner, Jr. Fund at Vanguard Charitable.

Mr Attia has equity in Eko, Inc. and has served as an advisor to AliveCor. Dr Kapa owns equity in Eko, Inc.; has served on the advisory boards of Myant, Inc. and Boston Scientific; and has received research grants form Abbott, Inc. and Toray, Inc.


Ko W-Y, Siontis KC, Attia ZI, et al. Detection of hypertrophic cardiomyopathy using a convolutional neural network-enabled electrocardiogram. J Am Coll Cardiol. 2020;75(7):722-733. doi: 10.1016/j.jacc.2019.12.030