PET image reconstruction using anatomical information through mutual information based priors

We propose a non-parametric method for incorporating information from co-registered anatomical images into PET image reconstruction through priors based on mutual information. Mutual information between feature vectors extracted from the anatomical and functional images is used as a priori information in a Bayesian framework for the reconstruction of the PET image. The computation of mutual information requires an estimate of the joint density of the two images, which is obtained by using the Parzen window method. Preconditioned conjugate gradient with a bent Armijo line-search is used to maximize the resulting posterior density. The performance of this method is compared with that using a Gaussian quadratic penalty, which does not use anatomical information. Simulation results are presented for PET and MR images generated from a slice of the Hoffman brain phantom. These indicate that mutual information based penalties can potentially provide superior quantitation compared to Gaussian quadratic penalties