An Investigation of the Required MR Bone Attenuation Correction for Quantitative Whole-Body PET/MR Imaging Using Clinical NaF PET/CT Studies

Tissue-classification-based attenuation correction strategies have been previously proposed to correct for bone attenuation in PET/MR imaging and simulated using computed tomography. However, the complication of voxel averaging uniquely associated with bone has not been considered explicitly in the past. This study investigated the effect of voxel averaging between bone and soft tissue in attenuation images and determined how accurately bone must be detected in MR images in order to perform acceptable attenuation correction of PET data by using CT-simulated attenuation correction. We found out that treating bone as soft tissue caused a mean quantification difference of -9.9% ± 5.5% in all 119 bone lesions. There were no significant differences between lesions in the pelvis and the vertebrae. The nominal difference in lesions in the ribs was significantly lower, likely due to the spatial misregistration between the emission and attenuation images. Interestingly, a non-monotonic relationship between the bone imaging ability and the absolute PET quantification accuracy was observed, with the minimal quantification difference achieved at a BVF around 40% for skull lesions (2.6% ± 2.1%), and 30% for non-skull lesions (1.4% ± 1.1%) and all lesions (1.5% ± 1.3%). This study established that a bone classification sensitivity of approximately 30% BVF is required in order for MR-based attenuation correction methods to achieve optimal quantification in whole-body PET/MR studies. For this purpose, higher bone imaging ability of MR may not be necessary.