Abstract
268
Objectives Iterative reconstruction can have a nonlinear response at low count densities (e.g., in dynamic or low-dose imaging) due to the positivity constraint in MLEM-based algorithms. The goal of this work was to investigate count-density-related biases in iterative list-mode reconstruction.
Methods List-mode (LM) data from a phantom with hot spheres were simulated for a range of count densities (100 kcts to 95 Mcts); a typical static FDG study has 20-40 Mcts/frame. Data were reconstructed using LM time-of-flight OSEM, and a global metric of total image counts was determined as a function of count density. The following aspects of reconstruction were investigated: chronological (time-ordered) subsets, a natural strategy for LM data vs. geometric (projection-angle) subsets used for binned data; sensitivity to the number of subsets; complementary reconstruction (CR), a technique recently proposed for low-count binned data where the image of a frame in a dynamic study is derived by subtracting the image of all data except that frame from the image of the full study.
Results LM TOF-OSEM of individual frames had a negative bias in image counts with decreasing count density for both chronological and geometric subsets; however, the magnitude of the bias exceeded 10% only for count levels rarely encountered in clinical research (<500 kcts). Using fewer chronological subsets reduced the bias below 8% for all count densities; the bias reduction was less with fewer geometric subsets. CR reduced the bias for geometric subsets to <2% for all count densities; for chronological subsets, a positive bias was observed.
Conclusions LM TOF-OSEM has a negative bias in total image counts with decreasing count density. Using fewer subsets for low count levels reduces this bias for chronological subsets but less so for geometric subsets. Complementary reconstruction significantly reduces the bias for geometric subsets but was not effective for chronological subsets.