Statistical atlases of anatomical structures have recently gained prominence in medical imaging research. Various applications include automatically segmenting anatomical structures, understanding population differences and creating patient specific models for intervention planning and guidance purposes. This thesis explores methods for constructing and validating stable statistical models of bone anatomical structures created from a training database of CT images, along with applications to 2D/3D registration and hip osteotomy procedure.;Our statistical model representation consists of a tetrahedral mesh for bone shape approximation and Bernstein polynomials for intensity. For the basic atlas construction, training images are aligned together using a gray scale deformable registration method and principal component analysis is performed on the mesh vertices and polynomial coefficients. An iterative bootstrapping technique to further stabilize the atlas variations is proposed. Atlases of male pelvis, female pelvis and female femur anatomy are constructed and validated using the proposed approach. A framework to construct and apply multi-component atlases of hip anatomy is presented. Finally, the application of statistical models in 2D/3D registration framework, osteotomy planning and intra-operative distortion correction is demonstrated.
展开▼
