Ultrasound registration for surgical navigation

In this thesis, we describe a method for registering intraoperative ultrasound (US) images of the bone surface to preoperative CT images. This registration method determines the spatial relationship between the relevant anatomy in the operative field and the preoperative plan. Thus, non-invasive ultrasound based registration can unlock the full potential of computer-assisted surgery by enabling the next generation of minimally-invasive surgical procedures.; Reliable segmentation of the bone surface from the ultrasound image is difficult problem due to the various noise, artifacts and imaging phenomenon of ultrasound imaging. The solution to this problem presented in this thesis avoids segmentation of the bone surface at the ultrasound image level, and instead combines three sources of partial information to segment regions that contain the bone surface reflection. The final segmentation process is performed in conjunction with the registration, using the 3D shape of the object to drive the final segmentation process.; Two methods for accomplishing the registration of US images to the preoperative CT are presented. The first method involves the preoperative construction of a surface model by segmentation of the bone surface from the CT image volume. The bone surface regions from the US images are then registered to the surface model using a variant of the iterative closest points (ICP) algorithm. The second US to CT registration method eliminates the construction of the surface model by using the current registration estimate to extract a region of the CT that corresponds to each US image. The bone surface from the US images and the corresponding CT images are then compared directly to determine the registration update transform. The theoretical formulation of both registration methods are presented along with experimental results to demonstrate the capabilities of ultrasound based registration.