Monte Carlo-based statistical SPECT reconstruction: influence of number of photon tracks

Quantitative accuracy of single-photon-emission computed tomography (SPECT) images is highly dependent on the photon scatter model used for image reconstruction. Monte Carlo simulation (MCS) is the most general tool for accurate modeling and correction of scatter, but calculations are notoriously slow. Recently, we proposed an efficient strategy for fully three-dimensional (3-D) statistical reconstruction using highly accelerated MCS as a forward projector. We use convolution forced detection (CFD) which accelerates convergence of reprojection by about two orders of magnitude. Here, we investigate how many photon histories during CFD-based MCS need to be calculated to: i) ensure that the noise in the reconstructed image does not increase markedly because of the stochastic nature of MCS reprojections; and ii) to determine the reconstruction time needed. To this end, phantom studies representing Tc-99m cardiac perfusion SPECT, were carried out. We generated reconstructions with different numbers of photon histories during MCS forward projection. Images and profiles were compared for reconstructions with a different number of photon tracks, for reconstructions from different noise realizations using a repeat measurement and for reconstructions with different random seeds for MCS reprojection. We found that 10⁵ photon histories per subset is sufficient to produce accurate images; more photons do not show visible improvement. This amount of photons corresponds to a typical reconstruction time below 5 min for a 64³ volume image on a dual-cpu PC (2.66 GHz), which is sufficiently short to apply such highly accurate reconstruction in clinical routine.