Registration-based motion compensation in 3D medical images

Abstract

Multi-modality medical images are widely used to provide complementary information to a clinician for diagnosis, treatment planning, and interventional procedures. They can also be used to provide supplementary information in alleviating the image degradation or improving the quality of target image. In those applications, accurate alignment via registration among multi-modal images or uni-modal images is imperative.This dissertation covers two kinds of registration-based motion compensation frameworks in 3D medical images: global motion compensation of computed tomography/magnetic resonance (CT/MR) images with respect to real-time ultrasound (US) images; and respiratory motion compensation in thoracic or abdominal positron emission tomography (PET) images.Most existing methods for 3D US-CT/MR registration require manual operation to determine initial registration transformation. In this dissertation, a novel feature-based registration algorithm between intra-operative 3D B-mode US and pre-operative 3D CT/MR images of the liver is proposed for automatic global motion compensation. In the proposed algorithm, to make the automation more robust to the US probe position and the degree of the initial displacement between two images, we utilize the geometric relation in vascular structures which is nearly invariant to the imaging modality. To effectively utilize the relation even in the non-rigid local deformation due to different respiratory phases, we proposed a robust graph-based edge matching algorithm. We also introduce a refinement process by jointly using vessel and liver surface information based on affine transformation to improve the registration accuracy. Experimental results on synthesized datasets and various clinical datasets show that registration error between anatomical features of US and CT images is about 3 mm on average even with a certain degree of local deformation as well as a large initial displacement.Meanwhile, for the respiratory motion co...