The aim of this paper is to present a complete methodology to perform the segmentation of a stack of MRI images. More precisely, we deal with (3D) of MRI images of mice brain. We first develop an automatic selection of the cerbellum area. Unadapted segmentation methods are mentionned. At last, we decided to use a second order decomposition method as a processing tool to perform denoising that does not erode contours.We end the segmentation a 3D Mumford-Shah method and present some results on the cerebellum volume estimates.