Application of the entropic mirror descent algorithm to TOF PET image reconstruction

In this work, we apply the theoretically convergent entropic mirror descent algorithm (EMDA) to the maximization of the list mode maximum likelihood objective function for time-of-flight (TOF) positron emission tomography (PET) image reconstruction. The algorithm also has a theoretically convergent ordered subsets implementation and is readily generalizable to priors. Here we demonstrate the algorithm on simulated TOF PET data and exhibit the improvement in initial convergence rate of the ordered subsets variant. We additionally demonstrate that the algorithm can be adapted to minimize the list mode likelihood with a total variation (TV) penalty term.