Accurate PET quantification requires attenuation correction (AC) for photon attenuation within the patient and within hardware components located between patient and de-tector. AC is a major challenge in hybrid PET/MR imaging, since standard MR images do not provide direct information on both patient and hardware attenuation. Conventional MR–based AC (MRAC) employed in clinical routine does not properly consider bone attenuation and entirely neglects attenuation of flexible hardware components such as MR–safe headphones. Both effects result in severe activity underestimation, especially in the brain region, making accurate PET quantification difficult. We have recently proposed two modifications of the maximum–likelihood reconstruction of attenuation and activity (MLAA) algorithm for non time–of–flight (TOF) PET/MR, simultaneously reconstructing attenuation and activity distributions from the PET emission data. MR–MLAA aims at improving patient attenuation estimation by incorporating MR–derived prior ex-pectations on the attenuation coefficients. The second algorithm, xMLAA, aims at estimating attenuation of flexible hardware components, without modifying the patient attenuation map. Both algorithms have been shown to significantly improve PET quantification compared to standard MRAC. In this work, MR–MLAA and xMLAA are combined to xMR–MLAA and applied to clinical PET/MR data of the head region. Compared to xMR–MLAA, conventional MRAC underestimates the average activity evaluated in the full brain by up to 15 %.