Optimization of image reconstruction protocol in neurological [۱۸F]FDG brain PET imaging using BGO-based Discovery IQ Scanner

Introduction: Since the Ordered Subset Expectation Maximization (OSEM) and Q.Clear algorithm each have advantages and disadvantages, we aimed to determine the optimal values of reconstruction protocols to achieve the best diagnostic parameters for the neurological PET brain images of  BGO-based PET/CT scanners. Methods: Images of point sources, as well as Hoffman and Carlson phantoms filled with [۱۸F]FDG radiopharmaceutical, were acquired using a PET/CT scanner. In OSEM, images were reconstructed with multiple iterations and subsets, applying ۳.۲ mm or ۶.۴ mm Gaussian filters, with PSF recovery enabled. For comparison, one reconstruction was done without PSF recovery using Iteration-Subset=۱۲–۱۲. In Q.Clear, β values from ۵۰ to ۵۰۰ in ۵۰-step increments were used for reconstruction. Parameters such as FWHM, COV and modified RC were evaluated. A cost function identified the best results, which were blindly assessed by two nuclear medicine experts for noise, contrast, and overall image quality.Results: Quantitatively, β=۵۰-۲۰۰ and Iteration-Subset=۲۰-۱۲ were the parameters whose Cost Function values were higher than Iteration-Subset =۱۲-۱۲, which was routinely used to reconstruct brain images in our center. Visual evaluations show that β=۲۰۰ has the lowest noise and the lowest contrast and evaluators gave the highest score for overall image quality to β=۲۰۰ and β=۱۵۰. This study has evaluated β=۲۰۰ and β=۱۵۰ as optimal for reconstructing brain images.Conclusions: This study investigated the different reconstruction algorithms to obtain the optimal parameters. The Q.clear algorithm with penalty function of β=۲۰۰ and β=۱۵۰ is recommended for brain neurological images of GE Healthcare PET/CT scanner.

Keywords:

PET/CT , Q.Clear , Bayesian penalized likelihood , Image quality , Optimization