Count-Regulated OSEM reconstruction

Block-iterative algorithms such as OSEM are widely used to accelerate tomographic image reconstruction in e.g. SPECT or PET. The speed-up factor of OSEM over MLEM is close to the number of subsets (NS). Significant additional speedup is achievable with novel pixel-based subset choices. However, a high NS can lead to undesirable noise amplification and unacceptable disappearance of activity. Here we introduce Count-Regulated OSEM (CROSEM): Given a maximum NS (NS^max), CROSEM automatically determines the voxel-wise reduction of the NS in low-activity voxels based on the estimated number of counts that originate from each voxel. CROSEM was compared to MLEM and OSEM for high-resolution multi-pinhole SPECT. SPECT simulations showed that for NS^max=128, CROSEM attained speed-up factors close to 128 in high-activity image regions and kept quantitative accuracy in low-activity regions comparable to MLEM. OSEM showed increasing quantification errors for an increasing NS. At equal contrast in high-activity regions, CROSEM exhibited lower noise in low-activity regions than MLEM.