Bayesian PET image reconstruction incorporating anato-functional joint entropy

We developed a maximum a posterior (MAP) reconstruction method for PET image reconstruction incorporating MR image information, with the joint entropy between the PET and MR image features serving as the prior. A non-parametric method was used to estimate the joint probability density (JPD) of the PET and MR images. The sampling rate for Parzen window estimation of the JPD was studied for both simulated phantom and clinical FDG PET brain images. Using realistic simulated PET and MR brain phantoms, the quantitative performance of the proposed algorithm was investigated. In particular, variations in the weighting factor on the MAP prior as well as the variance in the Parzen window were examined. Incorporation of the anatomical information via this technique was seen to noticeably improve the noise vs. bias tradeoff in various regions of interest.