Among the different methods proposed to perform attenuation correction in PET/MR, there are algorithms based on segmentation, on atlases, and on the simultaneous reconstruction of emission and attenuation data (Izquierdo-Garcia 2016, Chen 2017). Clinical scanners favor segmentation methods: a segmentation of the anatomical MR scan assigns each voxel to a tissue type and each tissue type to a predefined attenuation value. Current clinical scans use only a few tissue types and often misclasify bone as soft tissue (Martinez-Moller 2009).