Statistical method for source localization in MEG/EEG tomographic reconstruction problem

Characterizing the brain electromagnetic activity using magneto-encephalography (MEG) and/or electro-encephalography (EEG) requires solving an ill-posed inverse problem. This ill-posedness is due to spatial indetermination. Thus, restricting the number of the possible sources of the observed activity would improve the conditioning and facilitate the estimation of their amplitude. We developed a multivariate approach for estimating the spatial support of the electromagnetic cortical activity. Within the framework of a distributed source model and taking advantage of the linearity of the problem, we first performed an orthogonal decomposition of the space of possible source contributions. We then defined the subspace that explained the studied data best. The last step consisted in determining the sources that corresponded to this subspace. These sources defined brain activation areas associated to the observed data time window. The approach was validated using simulated MEG data.