A spatial bias in brain PET/MR exists compared with PET/CT, because of MR-based attenuation correction. We performed an evaluation among 4 institutions, 3 PET/MR systems, and 4 PET/CT systems using an anthropomorphic brain phantom, hypothesizing that the spatial bias would be minimized with CT-based attenuation correction (CTAC). Methods: The evaluation protocol was similar to the quantification of changes in neurologic PET studies. Regional analysis was conducted on 8 anatomic volumes of interest (VOIs) in gray matter on count-normalized, resolution-matched, coregistered data. On PET/MR systems, CTAC was applied as the reference method for attenuation correction. Results: With CTAC, visual and quantitative differences between PET/MR and PET/CT systems were minimized. Intersystem variation between institutions was 13.42% to −3.29% in all VOIs for PET/CT and 12.15% to −4.50% in all VOIs for PET/MR. PET/MR systems differed by 12.34% to −2.21%, 12.04% to −2.08%, and −1.77% to −5.37% when compared with a PET/CT system at each institution, and these differences were not significant (P $ 0.05). Conclusion: Visual and quantitative differences between PET/MR and PET/CT systems can be minimized by an accurate and standardized method of attenuation correction. If a method similar to CTAC can be implemented for brain PET/MRI, there is no reason why PET/MR should not perform as well as PET/CT. However, quantitative accuracy remains inconsistent between PET/MR and PET/CT systems, with MR-based attenuation correction (MRAC) suspected of being the main source of bias (3-5). MRAC remains a challenge, as MR images do not correspond to electron density and cannot be directly translated to m-values (6,7). Currently, conversion to m-values is performed via image segmentation (6-8).Although MRAC is feasible for whole-body imaging, large and spatially varying biases exist in brain PET/MR because of exclusion of bone in MRAC and incorrect segmentation of air cavities (3-5). These biases cannot be minimized merely by image postprocessing and must be compensated accordingly (9). An accurate, standardized attenuation correction method is needed to remove the differences.Currently, CT-based attenuation correction (CTAC) is an established standard (8) and is often considered the gold standard. Measured Hounsfield units (HUs) are converted to m-values by simple bilinear scaling (10-12). Therefore, a multicenter evaluation between PET/MR and PET/CT systems with CTAC and an anthropomorphic phantom would be highly desirable for determining image quantification in a controlled manner.Previous PET/MR investigations have been conducted with National Electrical Manufacturers Association whole-body or Hoffman phantoms, which do not model attenuation realistically and have not specifically addressed brain PET/MR (5,13,14). However, an anatomic brain phantom that has recently been developed models gray matter uptake and has a realistic head contour, including air spaces and the attenuation effect of bone (15,16). In this study, PET/MR and PET/CT syst...

No abstract