Automatic Prediction of the Origin in Outflow Tract Ventricular Arrhythmias with Machine Learning Combining Clinical Data and Electrocardiogram Analysis.

Identifying the site of origin in outflow tract ventricular arrhythmias (OTVAs) is crucial for the success of radiofrequency ablation procedures. Despite recent progress, this task remains challenging and too dependent on clinician's expertise, since origin estimation merely relied on visual inspection of the electrocardiogram (ECG). This study presents an automatic system to identify the ventricular origin in OTVA with machine learning algorithms. The system comprises two cascading classifier models that utilize raw electrocardiogram (ECG) signals, relevant ECG signal features, and clinical data. It was trained using data from four different databases. The final model achieved an accuracy of 95.45%. Furthermore, we identified specific regions in the ECG signal, such as the transition between the R and the S waves in V3 and V4 and the beginning of the QRS complex in V2, which are key when estimating the OTVA origin.