Artificial Intelligence for the Detection of Hypertrophic Cardiomyopathy From Standard Electrocardiogram

BACKGROUND: Hypertrophic cardiomyopathy (HCM) is often diagnosed late, increasing avoidable risk and delaying treatment. Artificial intelligence (AI) for 12-lead electrocardiograms (ECGs) may identify undetected HCM earlier. OBJECTIVES: This study aimed to evaluate the performance of an AI algorithm for HCM detection and identify predictors of correct classification. METHODS: Of 314 patients with cardiac magnetic resonance imaging (cMRI) for suspected HCM, 150 had analyzable ECGs and confirmed HCM by physician review of medical records and cMRI (ground truth). Eighty-three control patients without cardiomyopathy were included. A proprietary algorithm, Viz HCM (Viz.ai, Inc), labeled ECGs as HCM-positive or HCM-negative; the ECG closest to each patient's cMRI date was compared to ground truth. Diagnostic performance was evaluated by the area under the curve of the receiver-operating characteristic curve and at the prespecified threshold. Predictors of correct detection were determined by multivariable logistic regression for age, sex, race, maximal wall thickness, and hypertrophy subtype. RESULTS: The mean age of all 233 patients was 56 years, and 62% were male. The algorithm identified HCM with an area under the curve of 0.946 (95% CI: 0.916-0.970), sensitivity of 58% (95% CI: 50.0%-65.6%), and specificity of 100% (95% CI: 95.6%-100%). Apical subtype was a significant predictor of correct detection (adjusted OR: 4.71; 95% CI: 1.71-15.48; P = 0.005). In 9 of 28 patients with ECGs available at least 1 year prior to cMRI, the algorithm detected HCM 2.6 years (median) before clinical diagnosis. CONCLUSIONS: The AI ECG algorithm demonstrated highly specific HCM detection confirmed by cMRI and may improve early identification of unrecognized HCM.