Abstract : Magnetic resonance imaging is a grateful tool for observation of the human brain anatomy. In particular, the diversity of the parameters acquisition provides several views of the brain useful for the detection of brain tumours.
In the framework of the help of diagnosis, we study and propose an evidential segmentation scheme of multi-echoes MR brain images based on Demspter-Shafer theory. Considering each neighbor as an information source, we propose the use of a weighted spatial combination rule. It allows to consider each voxel in its spatial environment and leads to a real region segmentation. Applied to multi-echoes MR data, our process provides accurate segmentation of the brain and allows the tumours detection. Moreover, we study the conflict issued from the spatial combination process. We show the conflict is a new source of evidence which reflects the spatial organization of the data. In particular, this data can be used by specialists to soften the previous segmentation results.