Monotonic Algorithm For Joint Entropy-Based Anatomical Priors In Parametric Pet Image Reconstruction

In this paper we aim reconstruct kinetic parameters directly from dynamic PET sinograms. The reconstruction is performed by maximising the log-likelihood with a penalty term that is based on the joint-entropy (JE) with an anatomical prior. We used the surrogates by another group, combined with our own surrogates for the JE term. We developed 2 methods: the first method utilises the JE of the dynamic activity and the anatomical prior and the second one utilises the JE of the parameters and the anatomical priors. Results show the 2 approaches are monotonic and perform better than non-anatomically driven reconstruction in absence of inconsistencies between the activity and the anatomical prior. Also results suggest it is better to apply the prior on the dynamic activity rather than on the parameters.