Virtual Clinical Trials in PET Imaging for Improved Diagnosis of and Evaluation of Therapies for Cancer

Thesis (Ph.D.)--University of Washington, 2016-08

Positron emission tomography (PET) imaging is a diagnostic tool used to both quantify and verify the extent of disease, such as in cancer staging, and to monitor treatment response or measure disease progression. However, missed detection of tumors can lead to incorrect cancer staging and treatment selection, and uncertainty in quantifying tumor change can result in continuation of ineffective therapy. The objective of this work was to develop a virtual clinical trial methodology in PET imaging to enable evaluation of the impact of parameters on the final image analysis metrics. Numerous tools and techniques were combined, including radiotracer kinetic modeling, data generation simulations, image reconstruction algorithms, and human and model observer analyses. Improved understanding of parameters can be used to inform the design of prospective clinical trials to improve detection of tumors and measurement of response to therapy and ultimately improve cancer patient management. Lesion detectability using a new penalized likelihood (PL) reconstruction algorithm with a relative difference prior (RDP) was compared to that using OSEM, the standard clinical reconstruction algorithm. Due to the characteristics of the RDP, there was concern that low-contrast lesions were at risk of being smoothed into the background. Lesions in the liver and the lung were evaluated, and equivalent or improved detectability was demonstrated using the new algorithm. Optimum imaging time post-radiotracer injection for detection was investigated. Previous studies have shown that the tumor uptake increases for many hours past the standard imaging time of one-hour post injection, while uptake in the normal tissue decreases. Noise, however, also increases with time. It was hypothesized that there should be a time when the tumor signal-to-noise, and thus detectability, would be maximized. Lesion detectability was found to increase for several hours, indicating that delayed PET imaging may reveal low-conspicuity lesions that would have otherwise gone undetected. Finally, the uncertainties of static SUV and dynamic Ki metrics to measure change in tumor radiotracer uptake in response to therapy were characterized. For non-high-grade breast cancer tumors, it was found that Ki, while having higher variability, outperformed SUV in an ROC analysis of pre- and post-therapy parameter uncertainty distributions. Kinetic analysis, which accounts for metabolic state of the radiotracer, may better detect or enable earlier assessment of response to therapy, especially for low-uptake tumors.