UC Davis

Positron emission tomography (PET) imaging is a noninvasive imaging modality that provides in vivo visualization of biochemical processes in a living body through the use of radiotracers. With the latest state-of-the-art whole-body PET scanners using detector crystals of about 3 mm in size, the intrinsic spatial resolutions of PET scanners have been substantially improved. As a result, physiological motions, e.g., heart beating and respiratory motion, have become a limiting factor for PET spatial resolution in clinical practice. This thesis focuses on improving PET image quality by developing new methods for motion compensation and system modeling.Recently, deep learning has been successfully applied in various computer vision tasks, such as image segmentation, object detection, classification. First, we proposed a fully automatic data-driven gating approach using an unsupervised deep clustering network that exploits the autoencoder approach for respiratory gating. Second, we proposed a motion correction method of respiratory-gated PET images using a deep learning-based image registration framework. It does not require ground truth for training the network, which makes it very convenient to implement. We validated the proposed methods using simulation and clinical data and showed their ability to reduce the motion artifacts while utilizing all gated PET data. Furthermore, long axial field-of-view PET scanners provide high sensitivity and total-body coverage for accurate quantification of a wide range of physiological parameters in vivo using dynamic scans. We extended our proposed methods to obtain motion-frozen and motion-corrected total-body parametric images using deep learning-based data-driven gating and motion compensation techniques. Finally, we proposed a joint estimation framework incorporating deep learning-based image registration for motion estimation. In addition, a stochastic sampling method was developed to improve the spatial resolution of PET image reconstruction. This method enables efficient on-the-fly calculation of the detector response with DOI and we showed its potential to improve the spatial resolution efficiently for DOI-enabled high-resolution brain PET imaging.