Normalization of Monte Carlo PET data using GATE

The reconstruction of PET data without proper correction for non-uniform detection efficiency leads to severe artifacts in the image. This is also true for simulated data, particularly for Monte Carlo simulations that aim to realistically model the detectors. In the present study, we showed that the simulation with GATE software of the ECAT HR+, a block detector ring scanner, results in non negligible variations across the lines of response in the detection sensitivity. Like for real scanners, the line of response efficiency depends on its angle with the crystal axis and the intrinsic crystal efficiencies inside a block. To correct for such effects, we propose to use the same procedure as for real scanners: the simulation with GATE of normalization scans to estimate experimentally the different factors of a component based normalization model. With two distinct simulations, one for a 20 cm diameter cylindrical source and one for an annular source, we determined the detectors intrinsic efficiencies and the geometrical efficiencies. The results showed that none of the efficiency factors contributing to the component based normalization model can be ignored for Monte Carlo data. A high statistics acquisition of a uniform 30 cm diameter cylindrical source was simulated to evaluate the impact of normalization. The data were reconstructed analytically in 2D The non-uniformity was evaluated as the maximum amplitude variation in radial and axial profiles through the reconstructed image. Without normalization prior to the reconstruction, the non-uniformity index was 15% in the radial direction and 13% in the axial direction. Normalization reduced these numbers to respectively 0.7% and 1%. This clearly shows the benefit of using appropriate corrections for non-uniform detection efficiency for Monte Carlo simulated data and improves the realism of the simulated images.