Evaluation of model-based scatter correction accuracy using Monte Carlo calculated phantom inputs

3D scatter correction remains one of the leading challenges to clinical implementation of 3D PET. The Ollinger model-based 3D scatter correction method for PET was evaluated using Monte Carlo calculated input (single and multiple scatter) for two phantom configurations, a uniform 20 cm diameter cylinder, and an 8-box phantom (10 cm squares) with four radioactive, two air and two water compartments. GE PET Advance^TM 3D geometry was used. Phantoms were specified as contained within the field-of-view axially, within plus extending out 1 FOV axially on one side and within plus out 1 FOV on both sides. Ideal attenuation maps were used as generated from the simulation program. Simulation output was used as scatter correction input for trues+single-scatter and trues+total scatter. Parameters investigated for the scatter correction algorithm included: tail-scaling versus no tail-scaling (nulling error outside the object transaxially); 3 versus 5 iterations of the algorithm; inclusion versus exclusion of multiple-scatter and inclusion versus exclusion of extended axial FOV data (for extended phantoms). The 3 iteration scatter correction accuracy was not significantly different from that found with 5 iterations; single-scatter estimation is within 8% cif known Input for the 20 cm cylinder and within 20% with the 8-box phantom; when multiple scatter is included the correction accuracy 30% of an at-worst 120% correction. Further investigation of downsampling, the multiple-scatter model parameters, inclusion of extension data and the net effect of the correction on lesion detectability remains