On the convergence of iterative ordered-subset algorithms in small animal PET

The purpose of this study was to investigate the convergence of iterative ordered-subset algorithms in small animal PET studies. For routine practice, the most popular method is the OSEM which requires subset balance criterion for convergence. Another iterative ordered-subset algorithm, proven to be convergent, is the RAMLA which relies on a relaxation parameter to modulate the updating steps from iteration to iteration. An emerging ordered-subset and convergent algorithm is COSEM which elegantly integrates the complete data space to ensure algorithm convergence, and the ordered-subset to accelerate convergent rate. In this study, we will use clinical small animal PET data to evaluate the convergence performance of these three iterative ordered-subset algorithms under different conditions such as subset number and count rate. This study can also provide some guidelines for choosing appropriate algorithm with suitable parameter to achieve desirable convergence speed and accuracy.