Phantom and simulation study to determine image noise levels for a CT-based simulation platform of NaF PET uptake

Objectives Several acquisition and processing variables in sodium fluoride (NaF [F-18]) PET/CT imaging affect image noise. To characterize these factors, we are developing a patient-specific, CT-based simulation platform of NaF uptake. Fluoride uptake depends primarily on turnover rates of trabecular bone, which is readily identified on CT. Using acquired phantom data, this study’s purpose is to determine count-scaling factors for noise addition, which are scanner model dependent, that can equalize either sinogram counts or image noise between simulated and acquired data.

Methods Liver, spine, and interstitial compartments in the Anthropomorphic Torso Phantom were filled with CT contrast (2, 8, 0 mg I/cc, respectively) and F-18 (6, 48, 3 kBq/mL, respectively) to mimic the HU and radioactivity concentrations in organs observed on clinical NaF PET/CT. On a Discovery 690, CT and 45-min PET scans were acquired, and listmode data were rebinned into 10 frames of variable durations to encompass a range of total prompts: 10, 20, 30, 45, 60, 90, 120, 180, and 240 s. Data were reconstructed with MLEM of 40 iterations and FORE-FBP. Using the CT, NaF images were simulated by assigning radioactivity concentrations based on HU, forward projecting, scaling the sinogram counts, adding Poisson noise, and reconstructing with offline FBP and MLEM algorithms. ROI analysis was used to compare the simulated and acquired images.

Results Log-log relationships between sinogram counts and image noise were well-fit for both linear and iterative reconstructions of the simulated and acquired data. These relationships, which incorporate physical and reconstruction differences, yield conversion factors between acquired relative noise (n_a) and simulated relative noise (n_s) for sinograms of equal counts: FBP, n_s=0.905*n_a; MLEM of 40 iterations, n_s=0.956*n_a.

Conclusions Count-scaling factors were determined to equalize either sinogram counts or image noise between acquired PET data and CT-based simulations of NaF uptake.