ImprovedQuantificationAccuracyUsing Bayesian Penalized Likelihood BasedReconstruction on 68Ga PET-CT

162 Objectives: On the contrary to ordered subset expectation maximization (OSEM), Bayesian penalized likelihood (BPL) based PET reconstruction algorithm can reach the full convergence with improved image quality and reducing noise. Recent studies on BPL-based reconstruction mainly focus on 18F-FDG PET-CT, which reported to give higher SNR and standardized uptake values (SUV) compared with OSEM. Yet, the effect of BPL on image quality and quantification accuracy of 68Ga SiPM-based digital PET-CT has not been systematically reviewed. In this study, our aim was to evaluate the quantification accuracy of 68Ga SiPM-based digital PET-CT with BPL reconstruction method. Methods: For phantom study, a National Electrical Manufacturers Association (NEMA) image quality phantom was used with 6 spheres filled with 25.75kBq/mL gallium ions in a 1-to-4 ratio (background to sphere activity concentration). Phantom images were reconstructed using 4 different methods: (1) OSEM, (2) OSEM + time-of-flight (TOF), (3) OSEM + TOF + point spread function (PSF) and (4) BPL (Q.Clear, GE Healthcare, Milwaukee) In particular, images were also reconstructed using different penalization factor of BPL (β=100 - 1000, intervals of 100). The contrast recovery was then calculated.In clinical study, image data of 13 patients, with a total of 27 lymphatic lesions found within the abdominal area, were retrospectively reviewed and reconstructed using 4 different reconstruction methods - with the same protocol used in the phantom study. A blind visual image quality assessment was performed by two readers (Dr. X.Z. Hao and Dr. M. Liang, with 10 and 5 years of experience in nuclear medicine, respectively). SUVmax and SUVmean were calculated for all images reconstructed with different methods. For the statistical analysis, Cohen’s kappa coefficient (κ) was used to evaluate physician score consistency. Differences within the data were analyzed using the Kruskal-Wallis H test to evaluate the statistical significance for the phantom results and clinical data, with P < 0.05 considered as statistically significant. Results: Figure 1 has shown that, in phantom study, CR (%) increased gradually in the order from OSEM, TOF, TOF-PSF to BPL. As the sphere size increases, these CR% values also increase. Regardless of sphere size, images reconstructed using BPL method demonstrated higher CR% values than those of other algorithms (P < 0.05). The average CR% values calculated from BPL method can reach up to 36.65±3.71%, 43.09±1.06%, 57.72±1.19%, 64.27±2.49%, 70.51±3.61% and 80.80±0.81% (Fig.2). Table 1 shows the results of subjective image assessment for clinical images. Scores given from the 2 readers for general image quality, image sharpness demonstrated medium agreement, and lesion conspicuity demonstrated high agreement (κ= 0.655 for general image quality, 0.667 for image sharpness and 0.831 for lesion conspicuity). From both readers, beta value of 600 had the highest overall score. Values for SUVmax and SUVmean calculated using different reconstruction methods are shown in Figure 3.Increase trends of in the value of both quantification parameters were observed in the order from OSEM, OSEM + TOF, OSEM + TOF +PSF to BPL. In addition, statistical analysis demonstrated that SUVs determined from BPL method were significantly higher than those from other reconstruction methods (P < 0.05). Conclusions: Both phantom and clinical studies showed consistent results suggesting that the BPL-based reconstruction on the digital PET-CT demonstrated improved quantification accuracy with higher CR% approaching the true uptake and contrast ratio respectively.B value of 600 may be the optimal parameter for clinical 68Ga-PSMA PET-CT scan.