Deep learning–based ECG detection and phenotype classification of hypertrophic cardiomyopathy

Background Hypertrophic cardiomyopathy (HCM) is associated with increased risk of heart failure and sudden cardiac death, and early detection is essential for timely management. Although echocardiography is the standard diagnostic tool, it is resource-intensive and not always available. 12-lead electrocardiography (ECG) is rapid and widely accessible, making it a potential screening modality. We evaluated whether a deep learning model could detect HCM and differentiate the structural phenotypes using ECG data alone. Methods We retrospectively analyzed 568,252 ECGs from 184,287 adults at Ajou University Hospital between January 2020 and June 2023. Among patients with echocardiographically confirmed HCM, 1,713 ECGs from 450 patients (including 164 with apical HCM) obtained within ±30 days of diagnosis were labeled as HCM-positive. A Vision Transformer model was pre-trained on unlabeled ECGs and fine-tuned to classify three groups: non-HCM, typical HCM, and apical HCM. Negative controls were selected using age- and sex-matched sampling (1:20 ratio). Data were divided into training, validation, and test sets (7:1:2). Model performance was evaluated using area under the receiver operating characteristic curve (AUROC). Additional analysis was performed in patients with ECG-defined left ventricular hypertrophy (LVH). Results The model achieved an AUROC of 0.941 (95% CI 0.908–0.973) for detection of HCM, with accuracy 0.812, sensitivity 0.926, and specificity 0.806. Performance remained robust in the LVH subgroup (AUROC 0.917, 95% CI 0.876–0.958). For classification of apical versus typical HCM, the model achieved an AUROC of 0.869 (95% CI 0.798–0.939), supporting feasibility of phenotype discrimination using ECG alone. Conclusion A deep learning model trained on 12-lead ECGs accurately detects HCM and demonstrates potential for differentiating apical from typical HCM phenotypes. These findings support ECG-based artificial intelligence as a scalable screening strategy for early identification of HCM in routine clinical settings.