Localization of extended intracerebral current sources: Application to epilepsy

We propose a new method for localizing intracerebral current sources at the origin of epileptic spikes from non-invasive EEG/MEG data. This method was designed to account for three main constrains. First, most relevant spike generation models assume that sources are extended, i.e. spatially distributed over a focal or muti-focal area. Second, the background activity of the brain also contributes to EEG/MEG signals recorded during epileptic events. In this context, it can be seen as a penalizing Gaussian and spatially correlated noise. Third the array manifold is usually corrupted by errors due to the complexity of the conduction head volume. The proposed method is an adaptation of the well-established MUSIC method, that allows for the localization of Extended Sources (ExSo) assuming that all current dipoles comprised in the extended source are synchronous. In addition, we use Higher Order (HO) statistics, which are asymptotically insensitive to a Gaussian noise of unknown spatial coherence and which offer a greater robustness with respect to modeling errors. The method is called 2q-ExSo-MUSIC (q ges 2) as it combines the ExSo-MUSIC principle with the use of HO statistics. Using computer simulations of EEG signals, it is shown to highly increase the performance of classical MUSIC-like algorithms when physiologically relevant models for current sources and for volume conduction are considered.