Bayesian Estimation for Cardiac Activity Reconstruction Using Clinical Data

Bayesian estimation provides a unique strategy that can be used to reconstruct non-invasively the cardiac activity from torso recordings while incorporating prior knowledge about the underlying cardiac sources.This study investigates the accuracy of Bayesian estimation in localizing the origin of premature ventricular contractions (PVCs) for 10 patients.Surface recordings from 128 torso electrodes were captured before radiofrequency ablation.A training dataset containing simulated epicardial potentials, mimicking individual PVC, with 103 to 306 starting points per patient, was used as prior knowledge.All computations assumed homogeneous patient-specific torso models.Furthermore, the study investigates the effect of different training time intervals, ranging from 20% to 100% of the width of the QRS, on the accuracy of reconstruction in terms of localization error (LE).Results showed accuracy variations among patients and training intervals.The 100% training time interval had the smallest mean LE (31.0 16.1 mm) and median (25.3 mm), while the 60% interval had the highest mean LE (34.2 16.6 mm) and the 20% interval had the highest median of LE (30.0 mm).The study underscores Bayesian estimation's potential for cardiac activity reconstruction while investigating the impact of training time intervals on accuracy.