Qualitative and Quantitative Evaluation of Regularized Pet Image Reconstruction

579 Objectives Iterative reconstruction techniques such as OSEM are characterized by a tradeoff between bias and image noise while varying the number of iterations to reach convergence. Recently a penalized likelihood (regularized) reconstruction algorithm has been implemented on GE PET/CT scanners that reaches convergence without increasing image noise. The aim of this work is to evaluate the quantitative and qualitative performance of regularized reconstruction (RR) compared to OSEM of PET data with time of flight (TOF) and resolution recovery (PSF) using patient studies. Methods FDG PET scans of 14 patients (BMI range of 17-50) were reconstructed using (A) RR+TOF+PSF, (B) OSEM+TOF+PSF, (C) RR+PSF, and (D) OSEM+PSF. The OSEM reconstruction parameters were 3 iteration, 18 subsets, with 5mm Gaussian post filter. The resultant images were evaluated qualitatively by 2 radiologists on a scale of 1-4 for lesion conspicuity, artifacts and overall image noise. In addition, lesion SUVmax and SNR, defined as ratio of lesion SUVmax to standard deviation of SUV in a background ROI adjacent to the lesion, were calculated. In addition, liver uniformity was evaluated by measuring the standard deviation (STD) in a liver ROI. A total of 60 lesions were evaluated Results Qualitative results showed that algorithm A > B > C > D for lesion conspicuity, A = B > D > C for artifact reduction, and C > A > D > B for noise suppression. Lesion conspicuity of algorithm A was better by 18, 22 and 67% than algorithms B, C, and D respectively. Lesion SUVmax and SNR for algorithm A were on average 28%, 21%, 50% higher than algorithms B, C, and D respectively. The STD in the Liver was similar (p Conclusions RR results in better lesion conspicuity, higher SUVmax and SNR and lower noise content compared to other reconstructions. Artifact reduction is mainly due to TOF processing