Image quantification in high-resolution PET assessed with a new anthropomorphic brain phantom

Abstract

Choice of the PET scanner and image reconstruction parameters have significant impact in quantitative positron-emission tomography (PET). Hoffman phantom is probably the most widely used test object for assessing this impact in brain PET studies. In high-resolution PET, however, it's usability is questionable due to lesser partial-volume effect. Futhermore, Hoffman phantom is cylindrical and does not offer realistic attenuation effect for the skull. In the current work we used a novel brain phantom that was produced using a 3D-printer, and provides realistic head contour and skull attenuation effect. We scanned the phantom with latest generation whole-body PET/MR (Philips Ingenuity TF) and PET/CT (GE Discovery 690) scanners and in a brain dedicated high-resolution scanner (Siemens HRRT) to evaluate it's usability for intra- and inter-scanner comparisons with regard to PET brain imaging. In all scanners reconstruction algorithm choice and number of iterations had significant impact on anatomical gray matter ROI values. As compared to the HRRT, whole-body scanners showed 3% to 15% (Philips Ingenuity TF) and 0% to 5% (GE D690) negative biases in gray matter ROIs, when iterative reconstruction with high number of iterations but without resolution modeling was used. Whereas, low number of iterations in Philips Ingenuity yielded negative biases of 7% to 19%, but inclusion of resolution modeling in GE D690 yielded 19% to 7% higher values. In the HRRT count statistics related negative bias of up to 6% was seen, when iterative reconstruction without resolution modeling was used. We conclude that the new three-dimensional brain phantom is suitable for assessing the impact of reconstruction parameters both within and between scanners. However, the lack of ground truth values hampers the interpretation of the results, and furthermore, the small differences we saw between whole-body and brain-dedicated scanners might be due to limited resolution of the 3D-printing.