ISBN 9783843943512

978-3-8439-4351-2, Reihe Informatik

Jian Wang
Robust 2-D/3-D Registration for Real-Time Patient Motion Compensation

191 Seiten, Dissertation Universität Erlangen-Nürnberg (2020), Softcover, A5

Zusammenfassung / Abstract

In interventional radiology, preoperative 3-D images are usually fused with 2-D X-ray images. Fusion of 2-D/3-D images provides additional 3-D information during interventions and an accurate alignment between the 3-D and 2-D images is a prerequisite for fusion applications. This thesis focuses on a framework for dynamic, real-time and robust 2-D/3-D registration. The concept of depth layer image is introduced to set up correspondences between 3-D and 2-D images. It enables to estimate the 3-D motion by tracking of 2-D contour points in X-ray image sequences. To correctly describe the relationship between the 2-D motion and the 3-D motion, the point-to-plane correspondence (PPC) model is introduced, which provides efficient constraints on 3-D motion based on 2-D differential tracking. Based on the PPC model, a novel dynamic registration framework is proposed by involving 3-D update and direct 3-D/2-D misalignment measurement. This registration framework is capable of accurate and robust 2-D/3-D registration. It also shows its ability in ensuring the 3-D accuracy even for single-view registration. Real-time performance of the registration framework is achieved using the modern GPU. In order to enhance the robustness and efficiency of the registration framework, the PPC model is combined with the concept of depth layers initially proposed for motion compensation. In addition, the depth-encoded gradient projection rendering and structure tensor analysis are introduced, such that the apparent contour points are directly extracted from the depth layer images. This extension allows leaving out the surface point extraction from a 3-D volume and the apparent contour selection by thresholding the viewing angle. The extended approach requires short initialization time and fewer parameters, thus better meets the demand in clinical practice. The proposed dynamic registration framework is intuitive, generic, and suitable for both initial and dynamic registration scenarios.