A comparison of 3D scatter correction methods in a neuroPET tomograph

A comparison has been made between two procedures for scatter correction of 3D PET data acquired on a brain scanner (CTI/Siemens 953B). The methods are (i) dual energy window (DW), in which a weighted subtraction of data in a scatter window (200-380 keV) is made from the photopeak (380-850 keV) and (ii) convolution-subtraction (CS), in which a space-invariant kernel is derived from line source data. In multi-compartment phantoms, contrast ratios were restored to within about 6% of their true values with similar results for DW and CS. For both methods, good correction was observed in inactive water-filled compartments but residual counts were observed in air-filled cavities. Data from a patient injected with FDG was obtained by acquiring a 3D frame immediately after a standard 2D dynamic acquisition. Comparison of 2D and 3D images (scaled in kBq/ml using cross-calibration factors derived from phantoms) was made by plotting corresponding 8 mm square ROIs over the whole brain. The slopes/intercepts of these plots (ideally 1.0 and 0) were 0.97/-0.008 (DW), 0.95/-0.006 (CS) and 1.18/0.028 (uncorrected). Thus, overall similar results were obtained for the two methods although the noise in the corrected data was slightly higher for DW than for CS