Noise study in Monte Carlo estimated system matrix for OPET

The traditional method for estimating system matrices consists of simulating a number of annihilation events at each voxel position such that some error measured is minimized. However, axial sensitivity decreases as we move away from the center of the FOV and the number of detected coincidences is substantially smaller at the edge of the scanner than at the center. Therefore, the statistical error in computing should be increasingly larger for voxels at the edge of the image. To remedy in part this problem, we propose to estimate the system matrix using a number of events inversely proportional to plane sensitivity and compare its performance against the traditional approach by computing simple Figures-of-Merit. We simulate a noise cylinder phantom, a point-source phantom and a contrast phantom to measure noise, image resolution and contrast recovery, respectively, by image plane. Images are reconstructed using a ML-EM algorithm with system matrices built with the two methods. Simulation results show that noise is more uniform across the entire FOV when the matrix is estimated as proposed than when the traditional approach is used. Contrast recovery and resolution are not greatly improved.