Technique de parcellisation et de localisation des sources cerebrales a partir des signaux MEG

This work consists in the development of inverse problem resolution technique concerning the localisation of neuro-cerebral activity using MEG data. This technique, based on selective parcelling of cortical model sources, aims at providing a solution of the inverse problem after its regularisation.;The selective parcelling technique is based on two principles: sources pre-localisation and forming parcels using selected sources. For the sources pre-localisation, we used MSP (Multivariate Sources Prelocalisation) which helps estimating sources activity. Concerning sources selection we rely on hypothesis testing technique FDR (False Discovery Rate) which helps to define a selection threshold based on the sources pre-localisation data.;Based on results of selective parcelling technique, we then proceed to the inverse problem resolution using LCMV (Linearly Constrained Minimum Variance) technique which is adapted for Parcels spatial filtering. The results of using this technique of resolution show that we can achieve a good sensitivity and specify of activity sources localisation.;It is also important to mention that this technique is modular thus some parts of it can be reused independently of the others. For example, the selective parcelling technique can be combined with MEM (Maximum Entropy Method) technique (Amblard C 2004).