Prior Model Selection in Bayesian MAP Estimation-Based ECG Reconstruction

The inverse problem of electrocardiography (ECG) aims to reconstruct cardiac electrical activity using body surface potential measurements and a mathematical model of the body. However, this problem is ill-posed; therefore, it is essential to use prior information and regularize the solution to get an accurate solution. A statistical estimation has been applied to the inverse ECG problem with success, but a "good" a priori probability model is required. In this study, the Bayesian Maximum A Posteriori (MAP) estimation method is applied for solving the inverse ECG problem. Several prior models (training sets) are constructed, and the corresponding results are evaluated in terms of electrogram reconstruction, activation time estimation and pacing site localization accuracy. Our results showed that the training data consisting of beats from the 1
^st
or 2
^nd
neighbors of the test beat pacing nodes resulted in more successful results, implying that the prior models, including moderate amount and coverage of training data, might lead to an improved reconstruction of electrograms.