The impact of MR-guided PET attenuation correction on whole-body dynamic and parametric PET imaging

Objectives In this study, we investigate the impact of erroneous MRI-guided PET attenuation correction on dynamic and indirect as well as direct parametric whole-body (WB) PET data.

Methods Initially a series of realistic 4D PET XCAT simulation studies were conducted utilizing analytic projectors of Biograph mCT TOF PET scanner and WB FDG kinetics from literature to generate dynamic cardiac PET sinograms, sampled in frames of 45sec and at times corresponding to a 6-pass WB dynamic PET protocol. We focused on the impact of erroneously estimated attenuation factors (AFs) assuming perfect registration between CT- and MR-based attenuation maps. Initially, we extracted the CT AFs for all major organs from a dataset of 10 WB PET/CT clinical studies and assigned them to the respective XCAT organs to produce 10 cardiac XCAT CT attenuation correction maps (CT-AC) of realistic AF variance. Then, we generated 10 respective 3- and 5-class MR-based attenuation correction maps (3C-MRAC & 5C-MRAC). The original sinograms were first attenuated according to the 10 XCAT CT maps and then reconstructed using for AC either the CT-ACs or the 3- and 5-class MRACs. Finally, indirect and direct Patlak image generation methods were employed on the 3 datasets.

Results The average % bias observed between the 10 reference CT-ACs and respective 3C- and 5C-MRACs was from -20 to -25% for the bone and near bone regions and 5-10% for the lung regions. The same bias range was assessed for the dynamic 3C-MRAC PET images with a slight reduction by 1-2% in the regions with higher uptake. Finally a slightly reduced bias (2-4% less) was found for the 3C-MRAC indirect and direct Patlak images with no propagation to other regions. Finally, in the case of 5C-MRAC PET and Patlak images, the %bias in the bones became negligible.

Conclusions Standard 3C- and 5C-MRAC methods employed currently in the clinic can pose a quantitative challenge for future PET/MR dynamic and parametric WB studies with the % AC error being slightly reduced however in the case of Patlak.

Research Support This work was supported by the Swiss National Science Foundation under Grant SNSF 31003A-149957.