Bayesian estimates of error bounds for EEG source imaging

Given a set of electrical potential measurements at the surface of the head, localizing the sources of the electrical activity is an inherently ill-posed problem. Bayesian methods can be used to specify prior information to constrain the possible source solutions. We show that Bayesian analysis can also provide a means for characterizing system noise levels, estimating the "error bars" surrounding source localization results, and estimating the information about brain processes conveyed by dense sensor array electroencephalographic (EEG) recordings. This method is, in principal, applicable to any linear model of EEG or magnetoencephalographic (MEG) processes. A series of simulations demonstrated the internal consistency of our method, the robustness to noise levels, and the limitations of accurate source localization with large numbers of sources.