On the effect of relaxation in the convergence and quality of statistical image reconstruction for emission tomography using block-iterative algorithms

Relaxation is widely recognized as a useful tool for providing convergence in block-iterative algorithms [1], [2], [6]. In the present article we give new results on the convergence of RAMLA (Row Action Maximum Likelihood Algorithm) [2], filling some important theoretical gaps. Furthermore, because RAMLA and OS-EM (Ordered Subsets - Expectation Maximization) [4] are the algorithms for statistical reconstruction currently being used in commercial emission tomography scanners, we present a comparison between them from the viewpoint of a specific imaging task. Our experiments show the importance of relaxation to improve image quality.