Simulation-based evaluation of OSEM reconstruction bias on low activity PET data for the HRRT scanner

The ECAT High-Resolution Research Tomograph is a dedicated human brain positron emission tomography scanner. Due to gaps between its detector heads, 3D Ordinary Poisson OSEM algorithm is the reconstruction algorithm used in practice. For very low count statistics, it has been shown by several HRRT users groups that this algorithm converges to a biased solution. Unbiased quantification of low-activity regions is essential to measure accurately receptor-ligand kinetic parameters in brain receptor studies with reference tissue models. Using our simulation tool, we propose here to evaluate potential bias due to low-count statistics introduced by 3D OP-OSEM on binding potential (BP_ND). We also compare OP-OS NEG-ML, a scaled gradient ascent algorithm with an accelerate rate of convergence for low activity values. For simulated [¹¹C]-PE²I dynamic data (1 hour acquisition, frame durations ranging from 1 to 5 minutes), we observe that 3D OP-OSEM and OP-OS NEG-ML converge to the same activity for the specific region, and also to the same BP_ND. We show using simulated cylinder phantom that the observed bias for this specific protocol is mainly due to the partial volume effect. This does not prevent the existence of a low count statistics bias with OP-OSEM for other protocols.