MLAA–Performance for Reconstructing MR Hardware µ-Maps from non-TOF and TOF Emission Data

In PET/MR imaging, PET attenuation of MR hardware is currently either corrected using pre-defined CT based µ-maps or ignored for very low attenuating coils. Especially the latter case can lead to an underestimation of activity in the patient. To overcome this problem, this study investigates a maximum likelihood algorithm which updates the activity and attenuation distributions simultaneously from the acquired emission data (MLAA).Analytic simulations were performed to investigate the effect of the hardware (HW) and the potential of MLAA to mitigate for these effects. Both headphones (HP) and a multichannel MR torso coil were simulated. All data was then reconstructed using ordered subset MLAA (OS-MLAA) without any knowledge of HW attenuation, but with known "patient" attenuation (OS-iMLAA). The performance of OS-MLAA without any known phantom attenuation was also investigated. Simulations were carried out using time of flight (TOF) as well as non-TOF data. Based on previous studies, TOF emission data is expected to contain information that aids estimation of the attenuation.OS-MLAA reconstructed activity images showed a reduced mean error, standard deviation as well as root mean squared error compared to OSEM-reconstructed images for the same initial knowledge of attenuation. This behaviour was observed for non-TOF as well as for TOF data. Reconstructed µ-maps showed the main features of the HW that was used for reconstructions with and without known phantom attenuation. Their quantitative accuracy was not computed for this study.