Event-by-Event Motion Correction in Positron Emission Tomography: Development, Evaluation, and Applications

Positron Emission Tomography (PET) is a powerful metabolic and pharmacological imaging technique with a broad range of applications in Neuropsychiatry, Cardiology and Oncology. As the system resolution of PET imaging has improved, motion correction has become essential to eliminate image blurring and artifacts due to subject motion. Strategies of motion correction in PET include the frame-based methods and the event-by-event method. In the frame-based motion correction method, the scan data are divided into sub-frames, each of which is reconstructed individually, and registered to a reference orientation. Residual intraframe motion degrades image resolution. On the contrary, event-by-event motion correction transforms the location of every detected event to a reference orientation, thus eliminating intra-frame motion and giving the best theoretical accuracy. This algorithm has been implemented for the HART (High Resolution Research Tomograph) brain PET scanner in MOLAR (Motion-Compensation OSEM List-mode Algorithm for Resolution-Recovery Reconstruction).;This work develops, applies, and assesses the accuracy of both frame-based and event-byevent motion correction methods. Human and non-human primate (NHP) brain and whole-body PET studies were employed with 3 different scanners. This work is comprised of three projects, as described in the following paragraphs.;The first project compared the accuracy of frame-based motion correction and event-by-event motion correction methods in human brain PET imaging. Static and dynamic list-mode HRRT PET data were simulated using the forward projection model of MOLAR, incorporating a wide range of real subject head motion measured using the Polaris infra-red motion tracking system. Frame-based motion correction using image registration and the average of the externally tracked motion, and event-by-event motion correction were included in the image reconstruction using MOLAR. Image intensities in high-contrast regions of interest (ROI) and parameter estimates of volume of distribution and binding potential from tracer kinetic models were assessed to evaluate the accuracy of each motion correction method. The results showed that event-by-event motion correction can reliably correct for all head motions. With the attenuation map correctly aligned, frame-based motion correction produced ∼9% bias in ROI intensities, ∼5% in volume of distribution and∼10% in binding potential estimates for large motions (>5 mm).;The second project was aimed at developing, implementing, and evaluating both frame-based motion correction and event-by-event motion correction for awake NHP brain PET imaging on the microPET scanner FOCUS-220. For frame-based motion correction, a strategy for dividing PET list-mode data into sub-frames was developed with emphasis on the criteria in selecting the parameters in the sub-frame division algorithm. For event-by-event motion correction, the MOLAR platform was adapted for the FOCUS-220, requiring algorithms for detector efficiency normalization and calibration model for the count-rate dependent efficiency for the scanner. Both motion correction algorithms were applied to awake NHP PET studies with a GABAA-benzodiazepine receptor ligand [11C] fhunazenil. The reconstructed images were compared with those from a motion-free anesthetized study. The results showed that event-by-event motion correction noticeably eliminates image blurring due to intra-frame motion and improves statistics over frame-based motion correction method in awake NHP brain PET imaging.;The third project was aimed to develop event-by-event motion correction algorithm to correct for respiratory motion in whole-body PET studies on the Biograph mCT scanner, based on MOLAR. Application of MOLAR for the mCT required two novel algorithmic developments. First, in routine studies, the mCT collects list-mode data in 32-bit packets, where averaging of lines of response (LORs) reduced the number of LORs so that 32 bits are sufficient to address all sinogram bins. This degrades the spatial resolution. In this work, a probabilistic assignment of LOR positions (pLOR) was developed to address axial and transaxial LOR grouping in 32-bit data. Second, two approaches for 3D time-of-flight (TOF) scatter estimation were developed to accelerate the computationally intensive calculation. The proposed list-mode reconstruction algorithm was compared to the manufacturer's point spread function + time-of-flight (PSF+TOF) algorithm. Phantom, animal, and human studies demonstrated that MOLAR with pLOR gives slightly faster contrast recovery than the PSF+TOF algorithm that uses the average 32-bit LOR sinogram positioning. Moving phantom and a whole-body human study provided evidence that event-by-event motion correction reduces image blurring caused by respiratory motion.;The results from human brain studies, NHP brain studies, and whole-body studies indicate that, while frame-based motion correction methods may be suitable for human brain PET studies with small motion, the event-by-event motion correction method is clearly superior to the frame-based methods when large motion is present, e.g., in NHP brain studies and whole-body studies.