Population-based functional template priors for regularized PET reconstruction

We outline a possible method for exploiting population-based data for regularization in iterative PET reconstruction. Multi-modal and high-resolution mean shape templates are derived from a set of co-registered PET-MR images. The functional component of the template, representing the average radiotracer distribution among the images in the set, is used in a Bayesian reconstruction scheme for regularization of a given image. Unlike conventional anatomical-based priors, our proposed method makes no assumptions about relations between anatomy and function. Instead of regularizing based on differences between anatomy and function, we regularize based on differences between a mean functional image and a given functional image. Our proposed method outperforms both conventional MLEM and quadratic priors.