MAP estimation of PET images using prior anatomical information from MR scans

The authors present a maximum a posteriori (MAP) estimation algorithm that includes prior information, derived from registered magnetic resonance (MR) scans of the subject, in the reconstruction process. PET images are modeled as a number of homogeneous activity regions with abrupt changes in activity at the boundaries between these regions. These boundaries are likely to have close correspondence to anatomical boundaries in MR images. Thus, anatomical boundaries extracted from segmented MR images constitute the source of the prior information. The algorithm searches for smoothed PET images with discontinuities favored by both the PET data and the corresponding MR images. The potential performance of the method is tested using the Hoffman brain phantom and patient data