Heuristic modification of an anatomical Markov prior improves its performance

Including anatomical information during emission tomography reconstruction with resolution modeling can enhance the image quality. Often accurate segmentation of the anatomical image is required, being a major challenge for most applications. Recently, we studied a segmentation-free MAP algorithm proposed by Bowsher et al, that encourages similar activity in a selection of neighboring voxels that look most alike in the anatomical image. In an evaluation with Monte Carlo simulations, it was found to be very promising for both bias and noise reduction in 3D PET/MRI brain imaging, compared to MLEM and MAP algorithms with regular or anatomical priors. Here we study a small modification of the Bowsher algorithm to further improve its reconstruction capacities. Comparison between the two methods using the same brain phantom scan simulations indicated a further decrease in bias at the same noise level.