Comparative study on statistical image reconstruction combined with modification of the number of projections

An ordered subsets (OS) algorithm or a block iterative (BI) algorithm are useful methods to accelerate image reconstruction for ML-EM (maximum likelihood expectation maximization) or MAP-EM (maximum a posteriori expectation maximization). Generally the OS algorithm uses a fixed number of projections, so called "subset levels", and the recovered frequency component of a reconstructed image depends on the number of projections in a subset. In 1998 we proposed a new method which modified the number of projections in a subset for each iteration step in an OS-Bayesian reconstruction (OS-BR) algorithm to recover various frequency components in early iteration steps. In this paper we introduce our algorithm to the rescaled BI (RBI) version of the ML-EM algorithm (RBI-EM), and the RBI version of the MAP-EM algorithm (RBI-BR). And we evaluate the performance of proposed methods by using several simulation phantoms in comparison with those of ML-EM, MAP-EM, OS-EM and OS-BR.