Adaptive adjustment of the number of subsets during iterative image reconstruction

A common strategy to speed-up image reconstruction in tomography is to use subsets, i.e. only part of the data is used to compute the update, as for instance in the OSEM algorithm. However, most subset algorithms do not convergence or have a limit cycle. Different strategies to solve this problem exist, for instance using relaxation. The conceptually easiest mechanism is to reduce the number of subsets gradually during iterations. However, the optimal point to reduce the number of subsets is usually depends on many factors such as initialisation, the object itself, amount of noise etc. In this paper, we propose a simple scheme to automatically compute if the number of subsets is too large (or too small) and adjust the size of the data to consider in the next update automatically. The scheme is based on idea of computing two image updates corresponding to different parts of the data. A comparison of these updates then allows to see if the updates were sufficiently consistent or not. We illustrate this idea using 2 different subset algorithms: OSEM and OSSPS.