Clinical Application of a High Sensitivity BGO PET/CT Scanner: Effects of Acquisition Protocols and Reconstruction Parameters on Lesions Quantification

Aims: The aim of this retrospective study was to investigate SUVs variability with respect to lesion size, administered dose, and reconstruction algorithm. Background: SUV_max and SUV_peak are influenced by technical factors as count statistics and reconstruction algorithms. Objective: To fulfill the aim, we evaluated the SUVs variability with respect to lesion size, administered dose, and reconstruction algorithm (ordered - subset expectation _maximization plus point spread function option - OSEM+PSF, regularized Bayesian Penalized Likelihood - BPL) in a 5 - rings BGO PET/CT scanner. Methods: Discovery IQ scanner (GE Healthcare, Milwaukee, Wisconsin, US) was used for list mode acquisition of 25 FDG patients, 12 injected with 3.7 MBq/kg (Standard Dose protocol - SD) and 13 injected with 1.8 MBq/kg (Low Dose protocol - LD). Each acquisition was reconstructed at different time/FOV with both OSEM+PSF algorithm and BPL using seven different beta factors. SUVs were calculated in 70 lesions and analysed in function of time/FOV and Beta. Image quality was evaluated as a coefficient of variation of the liver (CV - liver). Results: SUVs were not considerably affected by time/FOV. However, SUVs were influenced by beta: differences were higher in small lesions (37% for SUV_max, 15% for SUV_peak) compared to larger ones (14% and 6%). CV - liver ranged from 6% with Beta-500 (LD and SD) to 13% with Beta- 200 (LD). CV - liver of BPL with Beta-350 (optimized for clinical practice in our institution) in LD was lower than CV - liver of OSEM+PSF in SD. Conclusion: When a high sensitivity 5 - rings BGO PET/CT scanner is used with the same reconstruction algorithm, quantification by means of SUV_max and SUV_peak is a robust standard compared to the activity and scan duration. However, both SUVs and image quality are influenced by reconstruction algorithms and the related parameters should be considered to obtain the best compromise between detectability, quantification, and noise.