Localisation des sources d'activité cérébrale à l'aide de la fusion multimodale EEG et MEG

Abstract

With the help of electroencephalography and magnetoencephalography measures, it is possible to localize the sources of brain activity in order to help diagnose diseases such as epilepsy. Even though the measures are acquired simultaneously, very few methods use the fusion of data for the sources reconstruction. The maximum entropy on the mean is an effective and flexible inverse problem technique where it’s possible to perform data fusion. In the present work, we propose a robust initialization technique for the maximum entropy on the mean method. It is shown that the perpendicularity constraint on the sources is a valid hypothesis in this framework. A new approach of the iterative MEM improves the source localisation. The work also proposes a novel clustering technique where the fusion of the information in the EEG and the MEG measures are taken into account. The results indicate that the empirical clustering technique improves the reconstruction of the sources with the MEM. Finally, a robust and valid framework was used throughout the present work.

Date	18 Mar 2010
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Jean-Marc Lina (Supervisor)