Source: Dissertation Abstracts International, Volume: 75-09(E), Section: B.

Adviser: Richard E. Carson.

Positron Emission Tomography (PET), as a functional imaging modality, has been widely used to provide quantitative measurements of biochemical and biological processes in vivo. Compared to the other medical imaging modalities, PET has lower spatial resolution. While the intrinsic spatial resolution is primarily determined by the design of PET systems, the effective spatial resolution is affected by many factors. These factors can be generally divided into two categories: algorithm-related factors, such as the imaging model and parameters in the reconstruction algorithms, and object-dependent factors, such as the spatial distribution of radioactivity and the statistics of the data. This dissertation investigates these resolution-related issues, and proposes and develops the corresponding improvements in the reconstruction algorithms to provide better resolution recovery. The contributions of this work are summarized in the following four projects.

The first project was aimed at developing an accurate imaging model for PET reconstruction. Resolution degradation in PET image reconstruction can be caused by inaccurate modeling of the physical factors in the acquisition process. Resolution modeling is a common technique that models the resolution-degrading factors in the system matrix. We developed a method of deriving the resolution kernels from Monte Carlo simulation and subsequently parameterizing the spatially-variant response functions. A thorough test was performed with this new model applied to two PET scanners - the HRRT and Focus-220. Some of the benefits of this new model, such as more uniform resolution recovery within the field of view (FOV) and higher image contrast at any given noise level, were verified by comparing with other models.

In the second project, we developed a practical method to regularize spatial resolution in the spatial and temporal domains for dynamic PET studies. In penalized likelihood reconstruction, due to the fact that system sensitivity and emission distribution are spatially and temporally variant, simple quadratic regularization does not produce uniform resolution over space or time. We extended the spatially-variant resolution regularization method of Tessler et al. to accommodate multi-frame reconstructions in 3D PET systems. The proposed regularization was compared with the original 2D method in HRRT simulation and phantom experiments. In single-frame reconstruction, our method reduced the axial dependence of image resolution. In multi-frame reconstruction, this method effectively achieved more consistent resolution recovery across time frames.

The third project investigated a subtle resolution-degrading factor in high-count PET scans. Most PET scanners with block detector designs suffer line-of-response misidentification effects called pile-up, where events tend to be mispositioned from the edge to the center of the blocks. For the HRRT with 2 detector layers (LSO and LYSO), there is an extra dimension of pile-up, i.e., inter-layer pile-up, which tends to push events to the LYSO layer at high count rates. Therefore, pileup results in error position information of coincidence events. This study was performed to characterize the resolution effects of this pile-up across the FOV by reconstructing simulated point sources at given levels of pile-up probability. In simulation, we have found a proportional relationship between pile-up probability and degradation of resolution. The conclusions were further verified in real phantom experiments. A potential correction strategy was proposed in image reconstruction procedure to provide better resolution recovery.

In the last project, we evaluated the low-count bias issue in our OSEM list-mode reconstruction platform (MOLAR). In both simulation and real phantom experiments, image bias was found to be small at for low-count images, unlike other OSEM algorithms. A relationship was found between bias and subsets number in OSEM reconstruction, i.e., a larger bias was observed in low-count reconstructions if the events were divided into larger numbers of subsets. This indicated that image bias and noise in ultra-low-count reconstructions with list-mode reconstruction can be potentially reduced by employing less subsets and more iterations.