3D/2D Registration of medical images

Abstract

The topic of this doctoral dissertation is registration of 3D medical images to corresponding projective 2D images, referred to as 3D/2D registration. There are numerous possible applications of 3D/2D registration in image-aided diagnosis and treatment. In most of the applications, 3D/2D registration provides the location and orientation of the structures in a preoperative 3D CT or MR image with respect to intraoperative 2D X-ray images. The proposed doctoral dissertation tries to find original solutions for the problem of multi-dimensional 3D/2D medical image registration by proposing three different 3D/2D registration methods. The dissertation also addresses the important problems of validation and comparison of different 3D/2D registration methods. The thesis describes the construction of two different “gold standard” image data sets and the corresponding evaluation methodology for the purpose of evaluation and comparison of proposed 3D/2D registration methods. The experimental results show that the three proposed registration methods are able to register a 3D image to two or more 2D images with accuracy that is sufficient for most orthopedic procedures. The proposed 3D/2D registration methods mostly differ in dependence of registration reliability on the initial registration provided by the user.