Magnetocardiography Inverse Estimation from a Realistic Ventricular Model Using Spatial Filter Methods: A Simulation Study

Abstract

We used a lead field normalization and measurement covariance matrix R into a weighted minimum-variance spatial filter (WMV). An inverse estimation, with an extended source, arranged at the surface of a realistic ventricular model was carried by WMV, weight-normalized minimum-norm (WMN), and minimum-variance spatial filter (MV) with and without noise. The performances of these spatial filters were evaluated using the estimation error and ratios of all positions with an estimation error below 2 cm. Moreover, the proper regularization parameter was determined from the estimation error. The results of the statistical analysis, while handling varied source positions, show that WMV has the best performance for magnetocardiography (MCG) extended source inverse estimation because it leads to less estimation error and is capable of stable inverse estimation, even at high noise levels. In other words, the combined lead field normalization, with the measurement covariance matrix R used in WMV, is a good choice for our MCG system.