Quantitative Evaluation of Reconstructed Image with Filtered Back Projection Bayes Method

We compared a Bayesian reconstruction method with conventional filtered back projection (FBP) method in positron emission tomography (PET). In the medical scene, PET scanning plays an important role for functional diagnosis such like measuring the cerebral glucose metabolisms. Tomographic image is reconstructed from the observed data, which is the recieved signals from administrated-radioactive substances existing in the target tissue, by a PET scanner. Various reconstruction methods have been proposed, and we focus on a Bayesian approach of our previous method, that is, introducing a Gaussian Markov random field (GMRF) as a prior and Gaussian observations as likelihood function. We evaluate reconstruction performances of our method in a region of interest (ROI) based approach and compare with the one of FBP method. In the result, we obtain our Bayesian approach might be able to suppress fluctuation of the observation noise.