3D MRI segmentation of brain structures

Some complex applications in medical imaging require to combine results coming from different numerical operators referred as to edges-and regions-detectors. Combining “static” contours and regions methods, named cooperative methods, takes advantage of these two compatible segmentation results. To improve success of segmentation, we propose the idea of combining region and active contour segmentations. In a first step, we use tools from mathematical morphology and region growing algorithm for the region segmentation in order to get the automated initialisation of the active contour model. In the second step, the physically-based active model considers the contour undergoing an elastic deformation as a set of masses linked by springs and converging to an equilibrium state. Segmentation results are shown on cerebellum, brain stem and hemispheres on 3D MRI data sets