A Machine Learning Approach for Idiopathic Ventricular Arrhythmia Source Classification

The study of idiopathic ventricular arrhythmias (IVAs), associated with premature ventricular contractions (PVCs) and ventricular tachycardia (VT), has a significant focus in clinical research as they may escalate into serious health conditions, including sudden cardiac arrest. These arrhythmias, which occur in the absence of structural heart disease, can often be traced to two primary anatomical regions: the right ventricular outflow tract (RVOT) and the left ventricular outflow tract (LVOT). Proper classification of these origins is critical for guiding treatment decisions, particularly when catheter ablation is considered. This study proposes a machine learning-based classification model for identifying IVA sources using 12-lead ECG data. Key features were extracted from ECG signals through multivariate variational mode decomposition (MVMD), phase-rectified signal averaging (PRSA), and mobility-complexity features. The extracted features demonstrated the distinguishing capability of origin of IVAs in RVOT and LVOT. Evaluated on a dataset of 334 patients, the proposed model used SVM and achieved an average sensitivity, specificity, accuracy, and F-score of 79.7%, 80.8%, 78.4%, and 77.1 %, respectively. These findings demonstrate the potential of automated ECG analysis for IVA source classification, with implications for enhancing IVA diagnostics and informing targeted ablation treatments.