Estimating the optimal iteration number in iterative reconstruction: A statistical approach

In MLEM, OSEM, and similar iterative algorithms, the contrast recovery improves with the iteration number, but image noise also increases with the iteration number, and the balance of these two opposite parameters is commonly left to an arbitrary choice of when to stop the iterative process. In the clinical practice, a fixed iteration number is a-priori selected and applied in all situations. In fact, the choice of the proper iteration number should be adaptive to different clinical applications, regions of interest, and most importantly, noise level or original data statistics. In this paper, we discuss an objective quantitative method to assess the optimal local iteration number that can provide a good balance between contrast recovery and low noise level if iterative algorithms are used for image reconstruction.