| Modern medical imaging techniques offer the opportunity to study human anatomy, function, and pathology in ways that were not possible before. In particular, the ability to perform imaging studies with large numbers of subjects in vivo has enabled studies of changes in brain structures during the time of development, aging, and pharmacological interventions. It has also created tremendous challenge for the image processing community to develop sophisticated and highly automated image analysis methods to identify and quantify subtle and spatially complex patterns of structural and functional changes in the brain. The work presented in this thesis is motivated by the need for an automated and accurate method for registering brain cortical surfaces to perform group analysis of the brain structure and function. The emphasis of the thesis is on efficiently modelling curve features on highly convoluted surfaces and on automated methods to extract the curves for deformable registration.; The thesis makes three major contributions. First, we developed a novel method to represent curves on highly convoluted surfaces via the new concept of anchor points, which are salient features along the curves. The introduction of this new concept enables us to efficiently represent the curves statistically, and makes the automatic extraction of these curves more accurate and more robust. Second, we developed a method for statistical shape modelling on the unit sphere. The method makes it possible to solve problems involving shapes on surfaces, where traditional active shape models in Euclidean spaces do not apply. Third, we developed a method to find weighted geodesic curves on convoluted surfaces incorporating global properties of the curves. The method finds weighted geodesic curves based not only on the local properties of the surfaces, but also on the global properties of the curves. |
