Automatic and interpretable prediction of the site of origin in outflow tract ventricular arrhythmias: machine...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Accurate identification of the site of origin (SOO) is crucial for successful radiofrequency ablation of outflow tract ventricular arrhythmias (OTVA).
• Current methods rely on subjective interpretation of electrocardiograms (ECGs) and physician expertise, often leading to inefficiencies.
• Existing computational and machine learning (ML) models face limitations in speed, interpretability, or data integration.

Purpose of the Study:

• To develop an automated ML strategy for predicting the ventricular origin of OTVA.
• To classify the specific ventricular origin (left/right outflow tract) using integrated ECG and clinical data.
• To explore the potential for detailed characterization of specific SOOs beyond simple ventricle laterality.

Main Methods:

• Supervised learning models were trained using ECG QRS complexes and clinical data from four distinct databases.
• Models were evaluated for their ability to classify ventricular origin (LVOT/RVOT) and characterize specific SOOs.
• Feature importance analysis focused on precordial leads (V1-V4) and QRS complex characteristics.

Main Results:

• The best ML model achieved an accuracy of 89% in predicting ventricular origin.
• Precordial leads V1-V4, particularly the R/S transition and QRS onset in V2, were identified as highly significant predictors.
• Unsupervised analysis indicated distinct clustering patterns for certain SOOs, suggesting identifiable electrophysiological signatures.

Conclusions:

• Automated ML prediction of OTVA origin using ECG and clinical data is feasible and effective.
• This approach offers a more objective and potentially faster alternative to current clinical methods.
• Identifying specific SOO patterns can further refine ablation strategies and improve patient outcomes.