Journées MAS de la SMAI
Modèles Spatiaux
Lille, 4 - 6 septembre 2006

Maureen CLERC

Spatio-temporal models in Brain Functional Imaging

All brain functional imaging modalities suffer from a trade-off in the spatio-temporal resolution which they can achieve: the most accurate modalities in space (Positron Emission Tomography, functional Magnetic Resonance Imaging) offer low time resolution, and conversely, the most accurate in time (Magneto- and Electro-encephalography abbreviated by MEG and EEG), are difficult to interpret in terms of spatial localization.

This talk will focus on the use of spatiotemporal models in electromagnetic source analysis.

Electromagnetic source analysis deals with the estimation of electrical activity in the cortex, a layer a few millimeters thick at the surface of the brain, from measurements of the electric potential (resp. magnetic field) outside the head with EEG (resp. MEG). The electrical activity is considered at a scale of a few square millimeters of cortex, and the sources can either be represented as linear combinations of isolated dipoles, or as surface distributions of dipolar current, oriented perpendicularily to the surface of the cortex.

The estimation of the sources producing the measured EEG or MEG is an ill-posed inverse problem. Although the uniqueness of the source distribution sometimes holds, the stability of the reconstruction is never guaranteed, and regularization is necessary. We will show how Markov Random Fields can be used to impose spatial neighborhood constraints. The spatio-temporal structure of the noise, in particular its covariance matrix, when available, must be incorporated in the model. A priori information on the source distribution is sometimes available, for instance through functional MRI measurements in similar experimental conditions. Bayesian models can incorporate such a priori knowledge.

Validating the source estimation results is a difficult task, because very little is known on the actual spatio-temporal behavior of the brain electrical activity. Permutation tests allow to evaluate the significance of the activation patterns.

Lastly, we will present some recent advances on identifying networks of brain activity, i.e. sets of brain regions dynamically interacting with each other.

• Sylvain Baillet and Line Garnero. A bayesian approach to introducing anatomo-functional priors in the EEG/MEG inverse problem. IEEE Transactions on Biomedical Engineering , 44(5):374--385, May 1997.
• D.Pantazis, T.E. Nichols, S.Baillet, and R.M. Leahy. \newblock A comparison of random field theory and permutation methods for the statistical analysis of MEG data. Neuroimage , 25:383--394, 2005.
• D.M. Schmidt, J.S. George, and C.C. Wood. Bayesian inference applied to the electromagnetic inverse problem. Human Brain Mapping , 7:195--212, 1999.


conférences plénières