Date of Award
Doctor of Philosophy (PhD)
Chair and Committee
X-ray computed tomography: CT) and positron emission tomography: PET) have become widely used imaging modalities for screening, diagnosis, and image-guided treatment planning. Along with the increased clinical use are increased demands for high image quality with reduced ionizing radiation dose to the patient. Despite their significantly high computational cost, statistical iterative reconstruction algorithms are known to reconstruct high-quality images from noisy tomographic datasets. The overall goal of this work is to design statistical reconstruction software for clinical x-ray CT scanners, and for a novel PET system that utilizes high-resolution detectors within the field of view of a whole-body PET scanner. The complex choices involved in the development and implementation of image reconstruction algorithms are fundamentally linked to the ways in which the data is acquired, and they require detailed knowledge of the various sources of signal degradation. Both of the imaging modalities investigated in this work have their own set of challenges. However, by utilizing an underlying statistical model for the measured data, we are able to use a common framework for this class of tomographic problems. We first present the details of a new fully 3D regularized statistical reconstruction algorithm for multislice helical CT. To reduce the computation time, the algorithm was carefully parallelized by identifying and taking advantage of the specific symmetry found in helical CT. Some basic image quality measures were evaluated using measured phantom and clinical datasets, and they indicate that our algorithm achieves comparable or superior performance over the fast analytical methods considered in this work. Next, we present our fully 3D reconstruction efforts for a high-resolution half-ring PET insert. We found that this unusual geometry requires extensive redevelopment of existing reconstruction methods in PET. We redesigned the major components of the data modeling process and incorporated them into our reconstruction algorithms. The algorithms were tested using simulated Monte Carlo data and phantom data acquired by a PET insert prototype system. Overall, we have developed new, computationally efficient methods to perform fully 3D statistical reconstructions on clinically-sized datasets.
Keesing, Daniel, "Development and Implementation of Fully 3D Statistical Image Reconstruction Algorithms for Helical CT and Half-Ring PET Insert System" (2009). All Theses and Dissertations (ETDs). 427.