In-Silico Data Based Machine Learning Technique Predicts Premature Ventricular Contraction Origin Coordinates

Premature ventricular contraction (PVC) can induce ventricular tachycardia or ventricular fibrillation. Drug-resistant PVCs can be cured by catheter ablation, but the accurate localization that this requires can be difficult and time-consuming. An accurate pre-procedural estimate of the origin could make the procedure more efficient. We propose a machine-learning method for accurate pre-procedural origin estimation. It uses a database of paced 12-lead ECGs with known pacing locations and presents its results on an imaging-based model of the patient. The method was tested using 7 realistic heart-torso models with hundreds of PVCs everywhere in the ventricles. We found that increasing the number of patients in the training database increased the accuracy of the predictions. The optimal number of pacing sites per patient in the training dataset was about 25, resulting in a prediction error around 15 mm. We conclude that our method gives a good indication to clinicians to efficiently start a pace-mapping during a catheter ablation procedure. It can be complemented with an intra-procedural method that uses the patient's own paced beats to refine the prediction.