Non-uniform Spatial Priors for Multi-Dipole Localization from MEG/EEG Data.

Localization of current dipoles from magneto/electro-encephalographic data is a key step in several applications, from basic neuroscience to pre-surgical evaluation of epileptic patients.SESAME is a Monte Carlo algorithm that can automatically localize an a priori unknown number of dipolar sources from M/EEG data, and provides a posterior probability map representing uncertainty on the source locations.SESAME has been shown to provide accurate localization in case of multi-dipole configurations. So far, SESAME has always been applied using a uniform prior distribution on the source location, corresponding to complete lack of information about the source location. However, in many practical contexts the experimenter (or clinician) might have some more or less vague information about where the sources could be.In this work, we investigate whether the use of non-uniform priors within SESAME can contribute to increasing the accuracy of source localization.We provide numerical results on simulated data, showing that the use of non-uniform priors can effectively increase the source localization accuracy when the prior distribution is correct (i.e., higher around the true source locations), without substantially worsening the performances when, as it may happen, the prior information is wrong.