| Description |
Positron emission tomography (PET) images can be reconstructed using a wide variety of techniques. Two aspects of image reconstruction are addressed in this thesis: the number of subsets used for the block-iterative ordered-subsets expectation-maximization (OSEM) reconstruction algorithm, and using smaller in-plane pixels. Both of these aspects of PET image reconstruction affect image quality. Although image quality in PET is difficult to quantify, it can be evaluated objectively using task-basked assessments such as lesion detection studies. The objective of this work was to evaluate both the effect of the number of OSEM subsets and pixel size on general oncologic PET lesion detection. Experimental phantom data were taken from the Utah PET Lesion Detection Database Resource, modeling whole-body oncologic 18F-FDG PET imaging of a 92kg patient. The data comprised multiple scans on a Biograph mCT time-of-flight (TOF) scanner, with up to 23 sources modeling lesions (diam. 6-16 mm) distributed throughout the phantom for each scan. Two observer studies were performed as part of this thesis. In the first study, images were reconstructed with maximum-likelihood expectation-maximization (MLEM) and with OSEM using 12 different numbers of subsets (i.e., 2-84 subsets). Localization receiver operating characteristics (LROC) analysis was applied using a mathematical observer. The probability of correct localization (PLOC) and the area under the LROC (ALROC) curve were used as figures-of merit in order to quantify lesion detection performance. The results demonstrated an overall decline in lesion detection performance as the number of subsets increased. This loss of image quality can be controlled using a moderate number of subsets (i.e., 12-14 or fewer). In the second study, images were reconstructed with 2.036 mm and 4.073 mm in-plane pixels. Similar LROC analysis methods were applied to determine lesion detection performance for each pixel size. The results of this study demonstrated that images with ~2 mm pixels provided higher lesion detection performance than those with ~4 mm pixels. The primary drawback of using smaller pixels (i.e. ~2 mm) was a 4-fold increase in reconstruction time and data storage requirements. Overall, this work demonstrated that reconstructing with moderate subsets or with smaller voxel sizes may provide important benefits for general PET cancer imaging. |
