Impact of out-of-field activity in MLAA estimation of lung attenuation for PET/MR

Attenuation correction of Positron Emission Tomography (PET) data using segmented Magnetic Resonance (MR) images is a promising approach to compensating for the lack of a transmission source in PET/MR and solving the problem of attenuation correction. However, the method is prone to errors in the choice of patient-individual lung attenuation coefficients (LAC), which are challenging to determine from MR data. Maximum-Likelihood reconstruction of Attenuation and Activity (MLAA) can be applied to reconstruct attenuation maps from PET emission data. We previously presented a constrained MLAA variant using segmentation of the lungs and tissue classification of the rest of the body to estimate mean LACs. In Monte-Carlo simulations restricted to the axial PET field of view (FOV), the mean LAC was found to be estimated with errors as low as five percent. In whole-body simulations, additional bias was observed and confirmed to be caused by uncorrected out-of-field (OOF) accidental coincidences. We have quantified the respective impact of uncorrected OOF scattered and OOF random coincidences. In this work, we discuss the prerequisites for clinical application of the proposed approach and propose potential improvements.