Out-of-field activity in the estimation of mean lung attenuation coefficient in PET/MR

In clinical PET/MR, photon attenuation is a source of potentially severe image artifacts. Correction approaches include those based on MR image segmentation, in which image voxels are classified and assigned predefined attenuation coefficients to obtain an attenuation map. In whole-body imaging, however, mean lung attenuation coefficients (LAC) can vary by a factor of 2, and the choice of inappropriate mean LAC can have significant impact on PET quantification. Previously, we proposed a method combining MR image segmentation, tissue classification and Maximum Likelihood reconstruction of Attenuation and Activity (MLAA) to estimate mean LAC values. In this work, we quantify the influence of out-of-field (OOF) accidental coincidences when acquiring data in a single bed position. We therefore carried out GATE simulations of realistic, whole-body activity and attenuation distributions derived from data of three patients. A bias of 15% was found and significantly reduced by removing OOF accidentals from our data, suggesting that OOF accidentals are the major contributor to the bias. We found approximately equal contributions from OOF scatter and OOF randoms, and present results after correction of the bias by rescaling of results. Results using temporal subsets suggest that 30-second acquisitions may be sufficient for estimation mean LAC with less than 5% uncertainty if mean bias can be corrected for.