Entirely automatic 3D MR brain analysis as a step in multimodal processing

Limited spatial resolution of positron emission tomography (PET) affects the quality of quantitative measurements. This difficulty can be overcome by the accurate spatial superimposition of magnetic resonance imaging (MRJ) and anatomical data bases. This paper deals mainly with the first step of this superimposition process. We describe an entirely automatic 3D analysis of brain MRJ data which first places the MRI set into an anatomical atlas co-ordinate system, then locates precisely the anterior and posterior commissures (AC, PC), and other anatomical structures and finally achieves a 3D segmentation of the brain. We use various image analysis technics : scene analysis, mathematical morphology and mean square fitting algorithms.