Analysis of the dependence of PET/CT quantification on iterative reconstruction parameters

Abstract

Although 18F-FDG PET/CT is used as a biomarker for predicting therapeutic response, there is a lack of standardized protocols for PET scans, in part, due to the variability in PET quantification with reconstruction parameters. Our objective is to facilitate PET standardization by assessing the variation of PET quantification with iterative reconstruction parameters. A NEMA IEC phantom with 6 spheres (ID: 37, 28, 22, 17, 13, and 10 mm) filled with 18F-water was scanned on a GE Discovery RX PET/CT scanner in 2D mode for different sphere-to-background ratios (SBR: 10, 5, and 3). PET data was reconstructed using 2D-OSEM with attenuation and scatter corrections for 66 combinations of iterations (IT: 1, 2, 3, 5, 7, 9, 15, 21, 30, 35, and 42) and subsets (SUB: 1, 3, 15, 21, 45, and 63). Regions-of-Interest (ROIs) were drawn over sphere and background regions on the central transverse slice on all reconstructions to calculate the mean, the maximum (max) and the standard deviation of ROI activity concentration (AC). The mean and max AC normalized to the true AC were analyzed as a function of IT, SUB, SBR and sphere size. The mean AC values converged to <10% of the true AC for all SUB as IT increased, it converged at a lower IT when using a higher SUB. The max AC values, although asymptotic in behavior, showed poorer convergence as IT increased and overestimated the true AC by 12-25% for different SUB. Similar findings for the mean and max AC with IT and SUB were observed for all SBRs and sphere sizes. Although the AC converged at high IT and SUB, the background noise also increased with both IT and SUB. The mean AC, max AC and background noise values exhibited <10% variability when the product of IT and SUB was held constant between reconstructions. PET/CT quantification using mean AC has substantially lower variability with iterative reconstruction parameters compared to max AC. Standardization of PET/CT activity quantification among scanners with different reconstruction parameters might be feasible when the product of IT and SUB is held constant. For a given combination of IT and SUB, the PET AC values were found to be independent of SBR.