Effect of Bayesian penalized-likelihood PET reconstruction condition on quantitation.

1601 Objectives: Quantitative analysis using standardized uptake value (SUV) is commonly used in oncological PET imaging. Recently, a Bayesian penalized-likelihood (PL) reconstruction algorithm has been developed to generate images with reduced noise and improved signal-to-noise ratio compared to conventional ordered subsets expectation maximization (OSEM) algorithm. However, it is not clear how the condition of PL reconstruction affects quantification of clinical PET. The purpose of this study was to investigate the effect of Bayesian penalized-likelihood (PL) PET reconstruction condition on quantitative parameters in FDG-PET/CT. Methods: In 21 patients suspected of having FDG-avid tumors, whole body PET/CT scans were performed at 60 min after iv injection of averagely 214.5 (range 130.9-330.2) MBq of FDG with a dedicated PET/CT scanner with 5-ring BGO detectors, 26 cm axial field of view. The PET images were reconstructed by PL and conventional OSEM (iteration 4, subsets 12) methods. In PL reconstruction, the noise reduction parameter (β) was determined as 300, by which the image quality was comparable with conventional OSEM. Maximum SUV (SUVmax) and highest mean SUV of 1cm3 spherical volume (SUVpeak) by PL and those by conventional OSEM were calculated for 31 FDG-avid tumors with the average size of 35.1mm (range 6-115mm). Metabolic tumor volume (MTV: volume of FDG-avid region greater than 40% of SUVmax), mean SUV in MTV (SUVmean) and total lesion glycolysis (TLG: MTV[asterisk]SUVmean) by PL and those by conventional OSEM were calculated and compared with each other. PL reconstructions with different β values (300, 400, 500, 600) were added and the parameters were compared with one another. Results: All parameters by PL showed strong correlation with those by conventional OSEM. SUVmax, SUVmean, SUVpeak and TLG by PL were significantly greater than those by conventional OSEM (p<0.01). However, the difference in MTV by the two reconstructions was not significant. Increasing β values decreased SUVmax, SUVmean and SUVpeak, but increased MTV. No significant trend was observed in TLG. Coefficient of variation (CV) of all parameters among different β values showed significant inverse correlation with the size of the tumor but did not show significant correlation with the injected dose. CV of SUVpeak was significantly smaller than that of other parameters (p<0.01). Conclusions: Although all quantitative parameters of PL method had strong correlation with those of conventional OSEM, significant differences in SUV parameters between the two methods should be noted for quantification of FDG-PET/CT. Although parameters fluctuate with β values, SUVpeak would be a reproducible parameter because of the smallest variance among different β values.