Evaluation of a method based on synthetic data inserted into raw data prior to reconstruction for the assessment of PET scanners

BACKGROUND: Performance assessment of positron emission tomography (PET) scanners is crucial to guide clinical practice with efficiency. Even though clinical data are the final target, their use to characterize systems response is constrained by the lack of ground truth. Phantom tests overcome this limitation by controlling the object of study, but remain simple and are not representative of patient complexity. The objective of this study is to evaluate the accuracy of a simulation method using synthetic spheres inserted into acquired raw data prior to reconstruction, simulating multiple scenarios in comparison with equivalent physical experiments. METHODS: We defined our experimental framework using the National Electrical Manufacturers Association NU-2 2018 Image Quality standard, but replaced the standard sphere set with more appropriate sizes (4, 5, 6, 8, 10 and 13 mm) better suited to current PET scanner performance. Four experiments, with different spheres-to-background ratios (2:1, 4:1, 6:1 and 8:1), were performed. An additional dataset was acquired with a radioactive background but no activity within the spheres (water only) to establish a baseline. Then, we artificially simulated radioactive spheres to reproduce other experiments using synthetic data inserted into the original sinogram. Images were reconstructed following standard guidelines using ordered subset expectation maximization algorithm along with a Bayesian penalized likelihood algorithm. We first visually compared experimental and simulated images. Afterward, we measured the activity concentration values into the spheres to calculate the mean and maximum recovery coefficients (RC(mean) and RC(max)) which we used in a quantitative analysis. RESULTS: No significant visual differences were identified between experimental and simulated series. Mann–Whitney U tests comparing simulated and experimental distributions showed no statistical differences for both RC(mean) (P value = 0.611) and RC(max) (P value = 0.720). Spearman tests revealed high correlation for RC(mean) (ρ = 0.974, P value < 0.001) and RC(max) (ρ = 0.974, P value < 0.001) between both datasets. From Bland–Altman plots, we highlighted slight shifts in RC(mean) and RC(max) of, respectively, 2.1 ± 16.9% and 3.3 ± 22.3%. CONCLUSIONS: We evaluated the efficiency of our hybrid method in faithfully mimicking practical situations producing satisfactory results compared to equivalent experimental data. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s40658-022-00496-6.