Abstract : This thesis comes within the scope of statistical shape modeling in 3D cerebral imaging.
In a first part, we propose a statistical shape model of cortical sulci. The model is built from a training population of sulci extracted from MRI volumes with a parametric representation. A coordinate system intrinsic to a sulcus shape is defined in order to align the training population, on which is then performed a principal components analysis. This statistical modeling is extended to a sulci graph in order to describe not only the morphological features of one sulcus, but also the relationships in terms of relative position and orientation between major sulci. The analysis we present is concerned with a reduced graph defined by a pair of sulci.
In a second part, three applications are considered. On the one hand, we take part to an evaluation project of inter-subjects brain registration methods. When performed on local landmarks, the statistical analysis provides a similarity measure between registered shapes, and thus provides a comparison criterion between methods. On the other hand, we exploit the statistical knowledge acquired by the sulci modeling in the context of anatomical and functional atlases building. More precisely, we propose a fusion scheme, local and non-linear, to register inter-subjects functional data (MEG dipoles) toward a single coordinate system linked to the anatomical model of cortical sulci. Experimented on a database of 18 subjects, this method has been shown to reduce the observed inter-individual functional variability. Last, the methodology proposed to model cortical sulci shape is applied to functional borders shape delimiting low-order visual areas.