Non-Invasive Approaches for Premature Ventricular Contractions Localization

Electrocardiographic Imaging (ECGI) has some limitations due to inaccuracies in some patients. Here, we explore 3 methods for the localization of premature ventricular contractions (PVCs): Tikhonov Regularization (TR), Bayesian Maximum A Posteriori Estimation (BMAP), and Multivariate Adaptive Regression Splines (MARS). A cohort of 5 patients with frequent PVCs was used, and the computations assumed both homogeneous and inhomogeneous torso models. Additionally, BMAP and MARS were computed using simulated prior knowledge of PVC endoepicardial potentials, originating either from the entire myocardium or a physiologically relevant area. The localization error (LE) was assessed as the Euclidean distance between the ECGI solution and the ground truth. BMAP method showed the lowest median LE with physiologically relevant prior knowledge for both torso models, outperforming TR. In contrast, the MARS method underperformed when applied to these pre-selected datasets. Overall, the results indicate that certain methods may improve accuracy under specific conditions, but further research is needed to explore their full potential and limitations in ECGI.