Improved Performance of Data-Adaptive Regression Framework Based on Multivariate Adaptive Regression Splines for Electrocardiographic Imaging

Noninvasive electrocardiographic imaging (ECGI) reconstructs cardiac electrical activity from body surface ECG.This study aims to 1) compare two ECGI methods, the traditional Tikhonov regularization with a nonparametric regression framework based on multivariate adaptative regression splines (MARS) using in-vivo experimental data, and 2) optimize how we use these data for training the MARS method.A regression model was trained with 184 body surface potentials and corresponding 239 unipolar electrograms (EGM) recorded from four anesthetized closed-chest pigs during sinus rhythm and pacing from various locations.A comparative analysis of four different training set compositions was performed on using a different number of beats.Performance of the methods was assessed by comparing estimated and true EGM and Activation Times (AT).The MARS-based method had better fidelity to the original EGMs compared to Tikhonov regularization when the training data available was sufficient, in number of beats and number of stimulus sites.The best compromise between performance and computing time was to train using 5 beats for each pacing site.