Evaluation of quantitative parameters of a new Bayesian penalized-likelihood reconstruction algorithm for PET, compared with conventional iterative reconstruction.

Objectives: Quantitative analysis using standardized uptake value (SUV) is commonly used in oncological FDG-PET imaging. Recently, a new 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 new PL reconstruction affects quantification of clinical FDG-PET. We evaluated characteristics of PL reconstruction in terms of quantitative parameters on clinical images compared with conventional OSEM.

Methods: A total of 29 whole body FDG-PET/CT scans were performed for oncological purpose with a dedicated PET/CT scanner with 5-ring BGO detectors, 26 cm axial field of view. 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 measured for 39 tumors with a diameter of 38.0±37.0mm. 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 measured for 31 tumors (on which MTV margin was definable). All parameters by PL and those by conventional OSEM were compared with each other.

Results: All parameters by PL showed strong correlation with those by conventional OSEM. SUVmax, SUVmean and SUVpeak by PL were significantly greater than those by conventional OSEM (p<0.001). However, the differences in MTV and TLG by the two reconstructions were not significant. The subtraction of SUVmax by conventional OSEM from that by PL showed significant negative correlation with the diameter of the tumor (r=-0.36, p=0.03). The same negative correlation was observed in SUVmean (r=-0.39, p=0.03) and MTV (r=-0.62, p=0.0002). However, no significant correlation was observed in SUVpeak and TLG.

Conclusion: Although all quantitative parameters of PL method had strong correlation with that of conventional OSEM, significant differences in SUV parameters between the two methods should be noted for quantification of PET/CT. The negative correlations between subtraction of SUV parameters by conventional OSEM from those by PL and the size of the tumor may suggest that the effect of PL is relatively greater in small lesions. Research Support: None