Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Extensive analysis of MR images of the brain can provide important informationabout neurological diseases and their progression. The study of subtle longitudinalchanges or anatomical variability between subjects requires accurate high-dimensionalnon-rigid image registration. This is not an easy task and to assist the registration,prior knowledge about shape variability in a population can be incorporated in theregistration algorithm. The goal of this master thesis is therefore to analyze andmodel deformations of groupwise registrations to develop an appropriate registrationprior that makes registration more robust. First, we look at possible ways to modeldeformations from atlas-to-image registrations. We explore a Gaussian model onthe deformations, we perform a PCA and we use a statistical measure based onthe Jacobian determinant of the deformation field. Both the Gaussian model andthe measure based on the Jacobian determinant are adequate methods and willbe used in a registration prior. First, the Gaussian registration prior is validatedwithin SPARC, a framework that simultaneously segments a heterogen dataset ofMR images, clusters them into different subgroups, builds up cluster-specific atlasesand registers each atlas to all the images. Registration results as well as the roleof the prior in the clustering is validated. The variance in the prior had a largeinfluence on both aspects and the incorporation of spatial variability in the varianceincreased the registration accuracy but slightly reduced the clustering performance.In a next step, we used the Jacobian determinant of the deformation field to builda stiffness map, which we incorporated in a registration prior. Using this prior, wetested the accuracy and the robustness of the registration. The use of the priorresulted in more consistent and robust deformation fields and with annealing onthe weight of the prior, accurate registration results where obtained. In the laststep of the thesis, we incorporated th

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)