Quantitative bias in PET/MR from attenuation correction and reconstruction: A comparison with PET and PET/CT with an anatomical brain phantom and Hoffman brain phantom

A comparison between PET/MR, PET/CT and PET was performed, to measure the bias due to attenuation coefficient assignment along with the effect of modifying the parameters in blob-RAMLA reconstruction algorithm. The measurements were performed by using an anatomical brain phantom with realistic head contour (Iida phantom) and a Hoffman brain phantom. The Iida phantom models the attenuation of the skull and uptake in the gray matter, with compartments for K₂HPO₄ solution and [F¹⁸]-FDG. The analysis consisted of comparison of 10 anatomical VOI regions and ratio images. Three blob parameters to maintain quantitative accuracy with improved image quality were evaluated. The error from misalignment and a location-specific bias was measured, for estimating the effect of attenuation. Additionally, attenuation correction was studied by comparing clinical MRAC with calculated MRAC derived from MR and measured attenuation maps derived from CTAC images. Modifying the blob parameters did not affect significantly to regional values on either of the phantoms; differences of 1-3 % in magnitude were measured. A misalignment of 2.53 mm was detected, producing asymmetry between hemispheres from 5 % to 19 %. After alignment correction, a location-specific bias ranging from 5 % to 17 % remained. The bias was reduced by taking into account the phantom skull either from calculated or measured attenuation correction; calculated attenuation introduced more bias. Selection of the RAMLA parameters accordingly did not affect quantification of the images. To improve the quantitative accuracy of the PET/MR systems, accurate tissue classification, attenuation coefficient assignment and alignment between PET and MRAC need to be accounted for. For testing and development of PET/MR instrumentation methods, novel phantoms fully compatible with PET/MR are needed.