Improved SAGE algorithm for PET image reconstruction using rescaled block-iterative method

Iterative reconstruction methods such as the expectation maximization maximum likelihood (EMML) method can be accelerated by using a rescaled block-iterative (RBI) algorithm. It was demonstrated that the space-alternating generalized expectation-maximization (SAGE) algorithm is superior to the EMML due to the following facts: (1) The hidden data spaces can be appropriately chosen and then be used in SAGE algorithm to speed up the convergence rate. (2) SAGE algorithm updates the parameters sequentially which makes its M-step to be treated more easily. In this paper, we present a novel algorithm that combines the RBI algorithm with SAGE algorithm. The convergence property of RBI-SAGE is discussed, and the image quality is assessed with mean absolute error and chi-square error. The experimental results show that the proposed method is more effective than the SAGE algorithm even if the projection data includes statistic noise.