high performance computing on graphics processing units: hgpu.org

As multiple sclerosis is known to cause atrophy and deformation in the brain, it also influences the shape and size of the corpus callosum. Longitudinal studies try to quantify these changes using medical image analysis techniques for measuring and analyzing the shape and size of a corpus callosum cross-sechtion embedded in a specially selected measurement plane. In this thesis, a framework has been implemented that automatically identifies and extracts the plane that contains the minimal cross-sectional area of the corpus callosum from a given MRI volume. The framework relies on deformable image registration for the segmentation and area calculation of the cross-section area of the corpus callosum embedded in a plane. Therefore, we used the free-form deformation transformation model, using B-splines, to chararacterize deformations based on a grid of control points. Computations that take place on a per-pixel basis have been transfered to the coarse grid of control points, leading to potential computational gains. To further improve the results of the registration process, a hierarchical multiresolution method is used that will increasingly refine the grid of control points. The whole registration process has also been accelerated by making use of the parallelization capabilities of a GPU. The registration process as well as the framework that uses deformable registration to identify and extract the plane containing the corpus callosum of minimal cross-sectional area, have been properly evaluated using synthetic and medical data.