PhD defence by Thomas Michael Alscher

Title
On a Collision Course
Abstract
Abstract Identifying trends, similarities and differences across multiple image series is a core challenge in various fields and requires finding the spatial transformations between those images to enable examination in a common frame of reference. This process, called image registration, is an important tool for extracting information from visual data. In this thesis, we focus on the medical application of image registrations, and specifically in 4DCT lung scans. The challenges in lung registration include finding complex non-linear, large deformation transformations that adhere to physical constraints as well as allowing for discontinuous motions taking place at the lung-thorax interface and between lung lobes. After an thorough introduction to the field of image registration and its foundation in physical deformation models, we develop a framework dealing with sliding motions via collision detection in the lung-thorax boundary with the application of adaptable Free-Form-Deformations. Apart from serving as a proof of concept, this work also compares the framework in a benchmark study on medical data to alterative solutions. This framework is used as the base to develop a flow-based generalization that elevates the method to diffeomorphic deformations. This step is accompanied by an extended medical evolution, with a special focus on diffeomorphims and their use in lung registration. Further extending the physics constrains to self-intersection problems, sliding motion in the lung fissures is explored. To that end a continuous lung model with lobe representations is developed and examined both in synthetic and medical experiments. Last, the framework design with supplemental computational analyses are presented in an implementation section. It elaborates on some lesser known core concepts of image registration and aims to provide an accessible approach to image registration. Concluding this thesis’ work is a discussion of the results in addition to prospective work inspired by our findings.
Supervisors
Principal Supervisor Sune Darkner
Co-Supervisor Christian Wong
Assessment Committee
Associate Professor, Melanie Ganz (chairman)
Professor, Anders Bjørnholm Dahl
Professor, Natasa Sladoje
Leader of defense: Melanie Ganz
For an electronic copy of the thesis, please visit the PhD Programme page.