Comparison of different approaches based on Monte Carlo methods to calculate the system matrix for small animal PET

This study focuses on Monte Carlo (MC) based techniques to calculate the system matrix (SM) for iterative image reconstruction of small animal PET data. For most PET scanners, the calculation of the SM requires extremely long simulation times due to the large number of image voxels, even when considering symmetries. Our goal is to determine in advance which simplifications can be done to accelerate the MC simulation without jeopardizing the accuracy of the SM. Our small animal scanner prototype was simulated using GATE. It consists of a continuous LSO crystal (42×42×10 mm3) attached to Flat Panel position sensitive photomultipliers. We investigated three different radioactivity distribution models to compute the SM elements for five representative voxel positions. The results show that the simulation time required to compute the whole SM and the accuracy of the SM depend on the model and on the voxel position. In the inner region of the field of view (FOV), the model with eight point sources distributed within each voxel yields the best trade-off between simulation time and SM accuracy. Whereas, for the peripheral region of the FOV, the model with one emission point at the center of the voxel yields the best compromise. The analysis was carried out with and without having depth-of-interaction (DOI) into account. The effect of the distribution model is more significant when DOI is considered.