Fully 3D system model estimation of OPET by Monte Carlo simulation

OPET is a combined optical and positron emission tomography (PET) scanner for small animal imaging, whose geometry presents several challenges for PET image reconstruction. The most important are the effects of depth-of-interaction, inter-crystal gaps, and angular sampling. Although these effects can be ameliorated through gantry rotations, the results are sub-optimal when reconstructing with filtered-back projection (FBP), because the system geometry is not modeled accurately. In this work, we present a study estimating the system model of the PET side of OPET with GATE, the Geant4 Application for PET and SPECT simulations, and the use of the model in iterative image reconstruction. At each image position, a 1 MBq voxel source irradiating isotropically in 3D was simulated. Coincidences were recorded for 10 seconds and the detected counts normalized and used as entries in the system matrix. Four transaxial and one axial image symmetries were exploited to reduce the number of simulations, but no assumptions were made on the model being axial or angularly independent. Resolution experiments were conducted by simulating a point source in warm background at various radial positions and images were reconstructed with the maximum likelihood expectation maximization (MLEM) algorithm. The simulations demonstrate that the use of the system model together with a statistical reconstruction produces images with improved and more uniform resolution across the field-of-view over FBP images.