Registration of multimodal three-dimensional medical images using points and surfaces

Registration of multimodal images makes it possible to combine different types of structural information (e.g., CT and MR images) and functional information (e.g., PET and SPECT images) for diagnosis and surgical planning. Registration of preoperative images with the physical space occupied by the patient is a fundamental step in interactive, image-guided surgery techniques. Of the many frameless methods that have been used to register medical images, point-based and surface-based methods appear to be the most useful for image-to-physical space registration. In this thesis we study a number of issues concerning these techniques: (1) We develop a technique for estimation of the localization accuracy of markers in three-dimensional images, and use it to analyze CT and MR image data obtained from phantom experiments and a clinical trial. (2) We investigate the accuracy of point-base medical image registration using implantable fiducial markers. We examine the theoretical accuracy of point-based registration, the experimental accuracy of a point-based neurosurgical system developed at Vanderbilt University using data obtained with a phantom, and the clinical accuracy of this system using data from a prospective clinical trial. (3) We compare the accuracy of surface-based registration of head image volumes using three popular techniques and two different surfaces (brain and skin). (4) We develop a technique for registration of three-dimensional image using multiple geometrical features, and demonstrate its efficacy by registering CT and MR head image volumes using various combinations of points and a surface. (5) We investigate the effect of geometrical distortion in MR images on the accuracy of registration of CT and MR head image volumes for both a point-based and a surface-based registration method.