A Deep Learning-based Classification Algorithm for the Origin of Premature Ventricular Contractions.

Catheter ablation is an effective and safe method for treating outflow tract ventricular arrhythmias. Preoperative preliminary localization of abnormal excitation origin is of great value in designing ablation strategies and improving the efficiency of ablation procedures. In this study, we collected surface electrocardiogram (ECG) data from 20 patients with premature ventricular contractions originating from the left ventricular outflow tract (LVOT) or right ventricular outflow tracts (RVOT) during catheter ablation procedures, with 10 cases for each group. We designed a two-stage model based on deep learning. This model automatically identifies the P, QRS, and T waves in the ECG and extracts ECG features to classify premature beats originating from LVOT or RVOT. To minimize model complexity, we compared different lead configurations and selected leads V2 and V3 as inputs to the model. The accuracy of the model in distinguishing between LVOT and RVOT origin beats was 98.46%, with a sensitivity of 97.94% and specificity of 99.23%. This study indicates that surface electrocardiogram can accurately classify the origin of PVC, laying a solid foundation for the preoperative design of catheter ablation strategies for these two types of premature ventricular contractions.