Electrocardiogram-based deep learning model for prediction of ventricular tachycardia in high-risk patients with hypertrophic cardiomyopathy

Background Sudden cardiac death (SCD) is the most feared complication of hypertrophic cardiomyopathy (HCM), mainly due to lethal ventricular arrhythmias. Objective This study aimed to create and evaluate a deep learning model that uses 12-lead ECG data to predict ventricular tachyarrhythmias in high-risk patients with HCM, which is a major precursor of SCD. Methods The deep learning model is tailored to predict future VT risk by using ECG data solely. In this retrospective multicenter study, high-risk HCM patients implanted with an implantable cardioverter-defibrillator (ICD) from 2015 to 2025 were enrolled. A total of 449 ECGs were used to develop a convolutional neural network (VGG18 model). K-Folding cross-validation with 10 folds was applied to enhance the robustness and generalizability of the established model. Results The VGG18 model demonstrated a good predictive performance, achieving an accuracy of 0.87, a sensitivity and specificity of 0.80, a precision of 0.87, an F1 score of 0.82, and an AUC of 0.93. In addition, the model exhibits effective learning and generalization without overfitting, as reflected by the increasing validation and training accuracy, coupled with the simultaneous decline in validation and training loss across epochs. Conclusion This model is a novel predictive tool capable of early addressing VT risk in a cohort of patients with HCM. The findings suggest that the model performs reliably and has the potential to support clinical decision-making in VT risk prediction, thereby helping to reduce adverse outcomes such as SCD at an earlier stage.