Machine Learning-Based Classification of Right and Left Ventricular Outflow Tract Premature Ventricular Contractions

Idiopathic premature ventricular contractions (PVCs) commonly originate from the right and left ventricular outflow tracts (RVOT and LVOT). The surface 12-lead electrocardiogram (ECG) is commonly used to differentiate the anatomic site of origin prior to catheter ablation. Machine learning methods can further enhance diagnostic accuracy.

In this study, 64 ECG-derived features, along with age and gender, were analyzed from 43 patients (65% males, median age 55.6, IQR 30 years), who underwent successful catheter ablation of outflow tract-originating idiopathic PVCs. The site of origin was depicted through 3-D electroanatomical mapping. A logistic regression model was trained to classify the site of origin as either RVOT or LVOT using repeated stratified five-fold cross-validation with five repeats. A model restricted to ECG features and four previously published ECG algorithms was evaluated for comparison.

Across the 25 held-out partitions, the model achieved an accuracy of 0.817 ± 0.143, precision of 0.888 ± 0.122, recall of 0.824 ± 0.206, F1-score of 0.836 ± 0.139 and ROC-AUC of 0.917 ± 0.094. Using the pooled out-of-fold predictions, accuracy was 0.860 (95% CI 0.744-0.953) and ROC-AUC 0.910 (95% CI 0.800-0.987). Age was the most discriminative feature, with increasing values favoring LVOT origin. The ECG-only model achieved an ROC-AUC of 0.853 (95% CI 0.721-0.958; ΔAUC +0.057, p = 0.104).

These findings suggest that machine learning models, based on ECG parameters and demographic characteristics, can classify outflow tract ventricular arrhythmias, supporting their use in automated differentiation and pre-procedural planning.