Acceleration of Monte Carlo SPECT simulation using convolution-based forced detection

Monte Carlo (MC) simulation is an established tool to calculate photon tran sport through tissue in Emission Computed Tomography (ECT). Since the first appearance of MC a large variety of variance reduction techniques (VRT) ha ve been introduced to speed up these notoriously slow simulations. One exam ple of a very effective and established VRT is known as forced detection (F D). In standard FD the path from the photon's scatter position to the camer a is chosen stochastically from the appropriate probability density functio n (PDF), modeling the distance-dependent detector response. In order to spe ed up MC we propose a convolution-based FD (CFD) which involves replacing t he sampling of the PDF by a convolution with a kernel which depends on the position of the scatter event. We validated CFD for parallel-hole Single Ph oton Emission Computed Tomography (SPECT) using a digital thorax phantom. C omparison of projections estimated with CFD and standard FD shows that both estimates converge to practically identical projections (maximum bias 0.9% of peak projection value), despite the slightly different photon paths use d in CFD and standard FD. Projections generated with CFD converge, however, to a noise-free projection up to one or two orders of magnitude faster, wh ich is extremely useful in many applications such as model-based image reco nstruction.