Whole-body direct 4D parametric PET imaging employing nested generalized Patlak expectation–maximization reconstruction. (6th July 2016)

• Record Type: Journal Article
• Authors: Karakatsanis, Nicolas A; Casey, Michael E; Lodge, Martin A; Rahmim, Arman; Zaidi, Habib
• Is Part Of: Physics in medicine & biology. Volume 61:Number 15(2016:Aug.)
• Page Start: 5456
• Page End: 5485
• Publication Date: 2016-07-06
• Subjects: PET -- parametric -- Patlak -- imaging -- whole-body -- quantification -- 4D reconstruction; Biophysics -- Periodicals; Medical physics -- Periodicals; 610.153
• DOI: 10.1088/0031-9155/61/15/5456
• Languages: English
• ISSNs: 0031-9155

Abstract

Abstract: Whole-body (WB) dynamic PET has recently demonstrated its potential in translating the quantitative benefits of parametric imaging to the clinic. Post-reconstruction standard Patlak (sPatlak) WB graphical analysis utilizes multi-bed multi-pass PET acquisition to produce quantitative WB images of the tracer influx rate K i as a complimentary metric to the semi-quantitative standardized uptake value (SUV). The resulting K i images may suffer from high noise due to the need for short acquisition frames. Meanwhile, a generalized Patlak (gPatlak) WB post-reconstruction method had been suggested to limit K i bias of sPatlak analysis at regions with non-negligible 18 F-FDG uptake reversibility; however, gPatlak analysis is non-linear and thus can further amplify noise. In the present study, we implemented, within the open-source software for tomographic image reconstruction platform, a clinically adoptable 4D WB reconstruction framework enabling efficient estimation of sPatlak and gPatlak images directly from dynamic multi-bed PET raw data with substantial noise reduction. Furthermore, we employed the optimization transfer methodology to accelerate 4D expectation–maximization (EM) convergence by nesting the fast image-based estimation of Patlak parameters within each iteration cycle of the slower projection-based estimation of dynamic PET images. The novel gPatlak 4D method was initialized from an optimized set of sPatlak ML-EM iterations to facilitate EM convergence. Initially, realistic simulations were conducted utilizing published 18 F-FDG kinetic parameters coupled with the XCAT phantom. Quantitative analyses illustrated enhanced K i target-to-background ratio (TBR) and especially contrast-to-noise ratio (CNR) performance for the 4D versus the indirect methods and static SUV. Furthermore, considerable convergence acceleration was observed for the nested algorithms involving 10–20 sub-iterations. Moreover, systematic reduction in K i % bias and improved TBR were observed for gPatlak versus sPatlak. Finally, validation on clinical WB dynamic data demonstrated the clinical feasibility and superior K i CNR performance for the proposed 4D framework compared to indirect Patlak and SUV imaging.