Optimization of Bayesian penalized likelihood reconstruction for 68Ga-prostate-specific membrane antigen-11 PET/computed tomography.

OBJECTIVE: The objective of this study is to determine the optimal β value for clinical use in digital 68Ga-prostate-specific membrane antigen (PSMA-11) PET/computed tomography (CT) imaging.

METHODS: 68Ga PSMA PET/CT of 21 patients with prostate cancer were reconstructed using block-sequential regularized expectation maximization (β value of 400-1600) and ordered subsets expectation maximization. Nine independent blinded readers evaluated each reconstruction for overall image quality, noise level and lesion detectability. Maximum standardized uptake value (SUVmax) of the most intense lesion, liver SUVmean and liver SUVSD were recorded. Lesions were then subdivided according to uptake and size; the SUVmax of these lesions were analyzed.

RESULTS: There is a statistically significant correlation between improvement in image quality and β value, with the best being β 1400. This trend was also seen in image noise (P < 0.001), with the least image noise reported with β 1400. Lesion detectability was not significantly different between the different β values (P =  0.6452). There was no statistically significant difference in SUVmax of the most intense lesion (P = 0.9966) and SUVmean of liver background between the different β values (P = 0.9999); however, the SUVSD of the liver background showed a clear trend, with the lowest with β 1400 (P = 0.0008). There was a decreasing trend observed in SUVmax when β values increased from 800 to 1400 for all four subgroups, and this decrease was greatest in small and low uptake lesions.

CONCLUSION: Bayesian penalized likelihood reconstruction algorithms improve image quality without affecting lesion detectability. A β value of 1400 is optimal.