Kirchmayer, B. (2021). Performance improvements and analysis of image reconstruction techniques in proton computed tomography [Diploma Thesis, Technische Universität Wien]. reposiTUm. https://doi.org/10.34726/hss.2021.94905
ion beam therapy; imaging; proton coputed tomograpy; pCT
en
Abstract:
The interest in proton beam therapy for the treatment of tumorous diseases hasincreased due to the unique energy deposition properties of protons, which allows to reduce dose exposition of healthy tissue. To fully utilize the potential of proton therapy, emphasis has to be put on treatment planning, which includes measuring the energy deposition per unit length (stopping power) of the tissue to be treated. Therefore, a 3D map of the stopping power of the affected tissue must be generated. Currently, this stopping power (SP) map is extrapolated from Hounsfield units (HU) as determined form x-ray computed tomography (CT) images, thus introducing conversion errors. By making use of proton computed tomography (pCT), the SP is measured directly. Difficulties arise as the particle trajectories deviate from straight lines due to multiple Coulomb scattering (MCS). Therefore the algorithms for image reconstruction commonly used in x-ray CT have to be adapted accordingly. For most implementations, this is done by introducing path estimates which account for the non-linear trajectory. Several methods have been introduced, including the most likely path (MLP) and cubic spline (CS) approximation. One of the main drawbacks introduced by the path estimates is the additional computational effort that has to be dealt with. As the reconstruction by itself is already computationally expensive, the need to calculate path estimates for pCT additionally increases computing time. One common approach to improve performance in scientific computing is to parallelize computations within the problem set. Since the computations in pCT image reconstruction are well suited to be executed in parallel, this approach has also gained momentum for application to pCT. In this thesis, the proton paths are estimated by a cubic spline, with the paths being used for generating improved radiographies as input for image reconstruction. The performance of the computations is increased by conducting parts of the calculations in parallel on a GPGPU (general purpose graphic processing unit). As the improved radiographies can be used as input for well established x-ray CT reconstruction methods, the software written in the process is used to expand the open source x-ray CT image reconstruction framework TIGRE [5]. Once implemented, the framework is used to perform image reconstruction for simulation data of several phantoms.
en
Additional information:
Abweichender Titel nach Übersetzung der Verfasserin/des Verfassers