Better conditioning the MEG/EEG inverse problem: the multivariate source prelocalization approach

The recently proposed multivariate source prelocalization (MSP) proved efficient and robust for restricting the solution space of the highly under-determined MEG/EEG inverse problem through the estimation of a probability-like coefficient of activation for each cortical area. It makes MSP a good candidate for initializing conventional inverse methods in order to improve electromagnetic source reconstruction. In this paper, we evaluate the benefit of MSP as a preprocessing step for a classical iterative weighted minimum norm algorithm, the FOCUSS approach, whose results highly rely upon its initialization. By using Monte Carlo simulations, we demonstrate the usefulness of such a coupling. Moreover, we quantify the benefits due on the one hand, to the reduction of the solution space and, on the other hand, to the explicit use as functional prior knowledge, of the activation probabilities inferred by MSP.