ABSTRACT
PURPOSE: In order to achieve comparability of image quality, harmonisation of PET system performance is imperative. In this study, prototype harmonisation criteria for PET brain studies were developed. METHODS: Twelve clinical PET/CT systems (4 GE, 4 Philips, 4 Siemens, including SiPM-based "digital" systems) were used to acquire 30-min PET scans of a Hoffman 3D Brain phantom filled with ~ 33 kBq·mL-1 [18F]FDG. Scan data were reconstructed using various reconstruction settings. The images were rigidly coregistered to a template (voxel size 1.17 × 1.17 × 2.00 mm3) onto which several volumes of interest (VOIs) were defined. Recovery coefficients (RC) and grey matter to white matter ratios (GMWMr) were derived for eroded (denoted in the text by subscript e) and non-eroded grey (GM) and white (WM) matter VOIs as well as a mid-phantom cold spot (VOIcold) and VOIs from the Hammers atlas. In addition, left-right hemisphere differences and voxel-by-voxel differences compared to a reference image were assessed. RESULTS: Systematic differences were observed for reconstructions with and without point-spread-function modelling (PSFON and PSFOFF, respectively). Normalising to image-derived activity, upper and lower limits ensuring image comparability were as follows: for PSFON, RCGMe = [0.97-1.01] and GMWMre = [3.51-3.91] for eroded VOI and RCGM = [0.78-0.83] and GMWMr = [1.77-2.06] for non-eroded VOI, and for PSFOFF, RCGMe = [0.92-0.99] and GMWMre = [3.14-3.68] for eroded VOI and RCGM = [0.75-0.81] and GMWMr = [1.72-1.95] for non-eroded VOI. CONCLUSIONS: To achieve inter-scanner comparability, we propose selecting reconstruction settings based on RCGMe and GMWMre as specified in "Results". These proposed standards should be tested prospectively to validate and/or refine the harmonisation criteria.
