Maximum likelihood reconstruction in fully 3D PET via the SAGE algorithm

The SAGE and ordered subsets algorithms have been proposed as fast methods to compute penalized maximum likelihood estimates in PET. The authors have implemented both for use in fully 3D PET and completed a preliminary evaluation. The technique used to compute the transition matrix is fully described. The evaluation suggests that the ordered subsets algorithm converges much faster than SAGE, but that it stops short of the optimal solution.