Automatic coordinate prediction of the exit of ventricular tachycardia from 12-lead electrocardiogram

Abstract

Automatic localization of the exit site of ventricular tachycardia (VT) can improve the efficiency and efficacy of catheter ablation. Because the exit site of the VT gives rise to its QRS complex on electrocardiogram (ECG), it is possible to build a predictive model to directly localize the exit of a VT from its 12-lead ECG. In previous works, prescribed features such as QRS integrals have been used to build such models. In this paper, we propose a deep network to automatically extract more discriminative features from QRS complex to localize the origin of ventricular activation. To improve the resolution of localization compared to previous works based on a small number of pre-defined segments, we localize the origin of ventricular activation as 3D coordinates. Model training and testing were performed on 12-lead ECG data of 1012 distinct pacing sites, collected from patients during routine pace-mapping procedures. Compared with the use of prescribed QRS-integral as an input feature, the presented deep model achieved an improvement of localization accuracy by approximately 4 millimeters (∼26%) on average.