Objectives: The penalized likelihood estimation reconstruction algorithm Q.Clear (GE Healthcare) allows for full convergence and edge preservation through a block sequential regularized expectation maximization technique. In this study the performance of Q.Clear was investigated for different penalization factors (β) with the aim to optimize its clinical use for four different tracers.

Methods: Q.Clear reconstructions with β values of 200, 400, 600 and 800 were compared to time-of-flight ordered subset expectation maximization (TF-OSEM) (3 iterations, 16 subsets and 5 mm Gaussian filter) with point spread function recovery. Clinical whole-body PET/CT (Discovery MI, GE Healthcare) scans with ⁶⁸Ga-DOTATOC, ¹⁸F-FDG, ¹¹C-acetate or ¹⁸F-fluoride were analyzed for level of noise in healthy liver tissue, signal to noise ratio (SNR), signal to background ratio (SBR) and maximum standardized uptake value (SUV_max). In addition, acquisition times per bed position and transaxial field of view (FOV) of the reconstructed images were varied. For each tracer, images from 10 patients were included, with a mean of 30 lesions per tracer. A spherical reference volume of interest (VOI) was placed in the liver and lesions were delineated employing a 41% threshold of the maximum voxel.

Results: The lowest levels of noise were reached with the highest beta factor resulting in the highest SNR, but this in turn gave the lowest SBR. Noise equivalence to OSEM was found with β 600 for ⁶⁸Ga-DOTATOC, ¹⁸F-FDG and ¹⁸F-fluoride, and β 400 for ¹¹C-acetate with a resulting significant increase of SUV_max (19.4%, 9.7%, 22.5% and 19.0% respectively) (P < 0.0001, paired t-test), SNR (22.1%, 22.6%, 9.5% and 33.6%) and SBR (19.5%, 11.7%, 21.3% and 18.5%) compared to OSEM. SNR decreased while SBR increased for all tracers when extending FOV from 500 to 700 mm, but only significantly for ¹⁸F-fluoride. Decreasing image acquisition time gave no statistical difference of SUV_max for ⁶⁸Ga-DOTATOC, ¹⁸F-fluoride (2 to 1.5 min) for any reconstruction method nor for ¹¹C-acetate (3 to 2 min) with β 蠅 400. Decreasing time for ¹⁸F-FDG (3 to 2 min) resulted in a change of optimal beta to β 800 in order to reach noise equivalence to OSEM along with maintaining a higher SNR than OSEM.

Conclusion: Images reconstructed by Q.Clear result in a tracer-dependent increase in tumour SUV_max values compared to OSEM at matched levels of noise, and an improved SNR. The optimal penalization factor, both in terms of noise-equivalence to OSEM and in terms of absolute SNR, is tracer dependent.