OS-EM and FBP reconstructions at low count rates: effect on 3D PET studies of [11C] WAY-100635.

11C-labeled neuroreceptor ligands frequently require long scan durations to quantify ligand-receptor binding. In this paper, we compare the accuracy of two three-dimensional (3D) positron emission tomography (PET) reconstructions: ordered-subset expectation-maximization (OS-EM) versus filtered backprojection (FBP) under low count rate conditions exhibited by 11C neuroreceptor studies. Data were obtained from a dynamic 11C phantom acquisition as well as six dynamic human [11C] WAY-100635 studies, all acquired in 3D mode using the EXACT HR+ PET scanner. Model-based scatter correction of the phantom datum was found to overcorrect in low count rate situations producing a negative bias in FBP reconstruction and a positive bias in OS-EM reconstruction, the OS-EM bias attributed to the non-negativity constraint of sinogram values. In the phantom OS-EM and FBP, reconstruction bias occurred at activities less than 25 Bq/cm3. In the human cerebellum, OS-EM deviated from FBP at activities less than 50 Bq/cm3. The total volume of distribution (VT), as determined from the metabolite corrected arterial input function and a two-tissue compartment kinetic model, was more sensitive to the positive bias of OS-EM than the negative bias of FBP at low count rates. To avoid reconstruction bias with 3D PET studies using the HR+, the scan duration should be limited so as to yield a final non-decay-corrected activity concentration of no less than 50 Bq/cm3. In neuroreceptor studies, if such a low count rate cannot be avoided, FBP reconstruction is preferable to OS-EM to estimate VT.