Combined statistical and geometrical 3D segmentation and measurement of brain structures

In this paper we present a novel method for three-dimensional segmentation and measurement of volumetric data based on the combination of statistical and geometrical information. The problem of shape representation of very complex three-dimensional structures, such as the brain cortex, is approached by combining the use of a discrete 3D mesh (the simplex mesh) with the construction of a smooth surface using triangular Gregory-Bézier patches. A Gaussian model for the tissues present in the image is adopted, and a classification procedure which also estimates and corrects for the bias field present in the MRI is used. Confidence bounds are produced for all the measurements, thus obtaining a distribution on the position of the surface segmenting the image as the output of the method. Performance is tested both on real data and simulations of MR volumes, which provide ground truth. The method is also compared with other existing techniques.