End-to-end deep learning–based electrocardiographic analysis for the detection of hypertrophic cardiomyopathy

Abstract Hypertrophic cardiomyopathy (HCM) shows heterogeneous clinical features and often leads to underdiagnosis and misdiagnosis, particularly when distinguishing it from other causes of left ventricular hypertrophy (LVH) and among HCM subtypes. Although electrocardiography (ECG) is widely available and sensitive, limited specificity constrains its standalone diagnostic value in HCM. We developed a deep learning–based model using standard 12-lead ECGs to simultaneously screen for LVH, distinguish HCM from phenocopies, and classify major HCM subtypes. This study innovatively emulates the clinical diagnostic workflow by establishing an end-to-end, hierarchical intelligent framework that encompasses the complete diagnostic pathway, integrating three key clinical decision-making stages into a unified model. Trained on 12,618 ECGs from 11,290 patients across three centers, the model showed strong performance in the internal testing set and maintained consistent performance in two independent external testing sets. Saliency analysis indicated that predictions relied on physiologically relevant ECG regions. This interpretable, ECG-based approach may support scalable HCM stratification and early phenotyping across diverse clinical settings.