Automatic segmentation of magnetic resonance images of the human brain via deformable models guided by probabilistic atlas of 3D salient points

Abstract

The magnetic resonance imaging (MRI) has become a fundamental tool in diagnosis and study of various diseases and syndromes of the central nervous system (CNS). Besides the systematic visual analysis of MR images, the neuroradiologist often need to measure the volume or analyze changes in the shape of certain brain structures to enable rapid and accurate diagnosis of a disease, or even to perform the follow up of a particular treatment. For that, a prior segmentation of the structures of interest is required. Usually, this task is done manually and because of this has several limitations. For this reason, many researchers have turned their efforts to the development of automatic techniques for the segmentation of tissues and brain structures in MR images. Among the approaches proposed in the literature, the ones based on geometric deformable models using probabilistic and topological atlases are among the techniques presenting the best results. This is because they allow the use of anatomical information inherently contained in the meshes during the segmentation process. However, a major difficulty applying geometric deformable models for medical image segmentation is the proper initial positioning of the model. Therefore, the main goal of this research project is to develop an automatic segmentation framework of some brain structures frequently used by neuroradiologists in the diagnosis of Alzheimer's disease and multiple sclerosis. This framework will be composed of three main parts: a 3D salient point detector, a probabilistic atlas of 3D salient points (built from a large MR image database) and a mechanism that will use both, the atlas and the salient point detector, for positioning the geometric deformable models. (AU)