Abstract
1836
Objectives The finite resolution of PET results in partial volume effects (PVE). PVE correction is a requisite for improvement of PET quantification of small structures. The aim of the present study was to experimentally assess the performance of the point spread function (PSF) reconstruction algorithm as implemented on the Ingenuity PET/CT scanner (Philips Healthcare, Cleveland, USA). Various parameter settings were explored to guarantee optimal use of this method.
Methods The NEMA image quality phantom and an anthropomorphic brain Hoffman phantom were used. The NEMA phantom was filled with [18F]FDG solutions of ~10 kBq∙mL-1 in spheres and ~1 kBq∙mL-1 in background. For the Hoffman phantom a solution of ~25 kBq∙mL-1 was used. Phantoms were scanned for 30 min on an Ingenuity PET/CT system and reconstructed with & without application of PSF. The number of PSF iterations (1-8) and regularization settings (0-12) were varied to optimize recovery performance.
Results For the NEMA phantom, the PSF algorithm provided a correction for PVE that improved with increasing number of iterations and larger regularization, but at the cost of increasing Gibbs/Ring artefacts, the latter resulting in large overshoots of the true activity concentration. The Gibbs artefact with 1 iteration and increasing regularization resulted in overshoots of 0.5 to 9% whereas with an increasing iteration and regularization=6 resulted in overshoots of 0.5 to 25%. In all cases 100% recovery for the smallest sphere was not achieved. Similarly, the Hoffman phantom data showed improved resolution recovery with increasing number of iterations and larger regularization. For both phantoms recovery improved with increasing image acquisition statistics.
Conclusions Use of the PSF algorithm on the Ingenuity PET/CT scanner results in improved resolution recovery, but at the cost of (sometimes severe) Gibbs artefacts. 1 iteration and regularization 6 provided reasonable contrast recovery without clearly notable Gibbs artefacts.