Automatic model selection for the 'best localization' of cardiac sources

In the present paper a simulation study on the influence of noise and source model on the accuracy of localization of the sources of biomagnetic fields is presented. Applying a statistical analysis (F test) to the localization results obtained by various models on the simulated maps calculated using different theoretical sources and different noise levels we were able to define a 'best localization' (BL) method. It allows an automatic determination of the particular source model able to represent in the best statistical way a specific field distribution obtaining the best source localization for that distribution. We applied this method to the localization of cardiac sources in the experimental maps of the magnetic field produced by isolated rabbit hearts completely immersed in a conductive medium. The results clearly indicate that the proposed method is very effective in determining the 'best localization' for every particular field distribution while the use of the same source model for every field map often produces a source localization completely in contrast with the position of cardiac structures of the isolated hearts.