Implementation and Parallelization of Cone-Beam Computed Tomography Back-projection on CPU and GPU
Bastian Weber
(Prof. Dr.-Ing. Volker Rodehorst, PD Dr. Andreas Jakoby)
Computed tomography (CT) is based on X-rays, which are able to penetrate solid materials, being attenuated in the process. This attenuation can be captured by a detector, giving an image of the attenuation properties of an object, which correlate to the shape and density of its material. CT is gaining more and more popularity for many applications in engineering and science. The most common cone-beam CT reconstruction algorithm for flat detectors is the Feldkamp-Davis-Kress (FDK) algorithm. The few known open-source implementations are very slow and not suited well for practical use. In this Master’s thesis we present a parallelized, fast and extensible open-source implementation of the FDK algorithm using C++ and CUDA, able to run on the CPU as well as on one or multiple GPUs. We describe the volumetric 3D reconstruction algorithm, discuss questions that arose and challenges we were facing and evaluate the results with regard to performance and quality.