Generalized 3D and 4D motion compensated whole-body PET image reconstruction employing nested EM deconvolution

Whole-body dynamic and parametric PET imaging has recently gained increased interest as a clinically feasible truly quantitative imaging solution for enhanced tumor detectability and treatment response monitoring in oncology. However, in comparison to static scans, dynamic PET acquisitions are longer, especially when extended to large axial field-of-view whole-body imaging, increasing the probability of voluntary (bulk) body motion. In this study we propose a generalized and novel motion-compensated PET image reconstruction (MCIR) framework to recover resolution from realistic motion-contaminated static (3D), dynamic (4D) and parametric PET images even without the need for gated acquisitions. The proposed algorithm has been designed for both single-bed and whole-body static and dynamic PET scans. It has been implemented in fully 3D space on STIR open-source platform by utilizing the concept of optimization transfer to efficiently compensate for motion at each tomographic expectation-maximization (EM) update through a nested Richardson-Lucy EM iterative deconvolution algorithm. The performance of the method, referred as nested RL-MCIR reconstruction, was evaluated on realistic 4D simulated anthropomorphic digital XCAT phantom data acquired with a clinically feasible whole-body dynamic PET protocol and contaminated with measured non-rigid motion from MRI scans of real human volunteers at multiple dynamic frames. Furthermore, in order to assess the impact of our method in whole-body PET parametric imaging, the reconstructed motion-corrected dynamic PET images were fitted with a multi-bed Patlak graphical analysis method to produce metabolic uptake rate (K_i parameter in Patlak model) images of highly quantitative value. Our quantitative Contrast-to-Noise (CNR) and noise vs. bias trade-off analysis results suggest considerable resolution enhancement in both dynamic and parametric motion-degraded whole-body PET images after applying nested RL-MCIR method, with- ut amplification of noise.