Surface Registration for Pharyngeal Radiation Treatment Planning Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 20, 2019
  • Zhao, Qingyu
    • Affiliation: College of Arts and Sciences, Department of Computer Science
  • Endoscopy is an in-body examination procedure that enables direct visualization of tumor spread on tissue surfaces. In the context of radiation treatment planning for throat cancer, there have been attempts to fuse this endoscopic information into the planning CT space for better tumor localization. One way to achieve this CT/Endoscope fusion is to first reconstruct a full 3D surface model from the endoscopic video and then register that surface into the CT space. These two steps both require an algorithm that can accurately register two or more surfaces. In this dissertation, I present a surface registration method I have developed, called Thin Shell Demons (TSD), for achieving the two goals mentioned above. There are two key aspects in TSD: geometry and mechanics. First, I develop a novel surface geometric feature descriptor based on multi-scale curvatures that can accurately capture local shape information. I show that the descriptor can be effectively used in TSD and other surface registration frameworks, such as spectral graph matching. Second, I adopt a physical thin shell model in TSD to produce realistic surface deformation in the registration process. I also extend this physical model for orthotropic thin shells and propose a probabilistic framework to learn orthotropic stiffness parameters from a group of known deformations. The anisotropic stiffness learning opens up a new perspective to shape analysis and allows more accurate surface deformation and registration in the TSD framework. Finally, I show that TSD can also be extended into a novel groupwise registration framework. The advantages of Thin Shell Demons allow us to build a complete 3D model of the throat, called an endoscopogram, from a group of single-frame-based reconstructions. It also allows us to register an endoscopogram to a CT segmentation surface, thereby allowing information transfer for treatment planning.
Date of publication
Resource type
Rights statement
  • In Copyright
  • Niethammer, Marc
  • Marron, James Stephen
  • Pizer, Stephen M.
  • Alterovitz, Ron
  • Rosenman, Julian
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill Graduate School
Graduation year
  • 2017

This work has no parents.