Dynamic PET imaging usually involves the collection of a series of frames of sinogram data over contiguous short-time intervals. Since scans with shorter acquisition time are usually very noisy, iterative image reconstruction can not ensure for both stable and fast convergence. The more recently row-action maximum-likelihood algorithm (RAMLA) allows sequences of orthogonal projections and a relaxation parameter to control updating of the log-likelihood objective for sure convergence. For noisy and low-count sinogram, larger subset sizes with appropriate choice of relaxation scheme can be combined to achieve fast convergence. In this work, we investigate RAMLA with fixed relaxation parameter and with subset-dependent relaxation parameter of Tanaka and Kudo for microPET dynamic scans. The experimental results have indicated the RAMLA with subset-dependent relaxation parameter can reach the plateau of the likelihood function faster than RMLA with fixed relaxation. The OSEM algorithm with the same subset size fails to converge for such a low-count case. The smaller subset of OSEM is used, but the likelihood function converges to a sub-optimal solution. The results suggest that the RAMLA with subset-dependent relaxation scheme is suitable for low-count dynamic PET imaging using microPET.
