|Nowadays almost all commercial PET systems are PET/CT scanners. Also on the horizontal are combined PET/MR scanners. PET/CT systems provide both the anatomical and functional images during a single imaging session. The combined anatomical and functional images have improved the ability to distinguish pathology from normal uptake and to precisely localize the abnormal loci. However, image reconstruction methods have not been able to take advantage of the self-registered anatomical information. The idea of using anatomical prior information in PET reconstruction has been proposed since early 1990s, but challenges remain because of the signal and position mismatches caused by the difference in the contrast mechanism and/or physiological motion. In our work, we develop advanced image reconstruction methods for PET to address this problem. In (Liao_MIC07), we proposed a Bayesian approach where we used level set functions to describe the anatomical boundaries from CT and also to track the evolution of functional boundaries in PET. We evaluated our method using a digital phantom obtained from a PET/CT scan of a mouse and demonstrated that it provides results that are locally smooth but have sharp functional boundaries. The ROI quantification study also indicated that our method achieves less bias than the existing algorithms at the same noise level. We are extending our results to clinical PET/CT studies.