Efficient simulations of iodine 131 SPECT scans using GATE

Abstract

To speed up Monte Carlo simulations of Single Photon Emission Computed Tomography (SPECT) scans using iodine-131, a tabulated modeling of the detector response has been incorporated within the GATE simulation toolkit, based on the use of the Angular Response Function (ARF). In this work, we validate the ARF methodology within GATE for I-131 simulations and demonstrate the practical feasibility of the simulation of I-131 SPECT patient acquisitions. The Siemens Symbia T equipped with a high energy collimator was considered. Planar acquisitions of I-131 point and plane sources in air were simulated using GATE without (sGATE) and with the ARF model (ARF-GATE). Profiles through the projections and root mean square differences (RMSD) between ARF-GATE and sGATE projections were calculated. The statistical distributions of the simulated projections were also investigated. A I-131 Lipiocis <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">?</sup> SPECT scan was also simulated and the simulated and acquired projections were compared. Profiles across the ARF-GATE et sGATE projections agreed well for the point and plane sources, with RMSD of 9%, similar to those obtained between two independent sGATE projections. The ARF-GATE and sGATE projections were Poisson distributed. About 36 times less photons were needed with ARF-GATE than with sGATE to get images of equivalent statistical quality for the plane source. Overall, ARF-GATE produced images indistinguishable from the sGATE images in 140 less time. For the patient simulations, simulated projections of visually comparable quality as acquired projections were obtained in 100,000 s. The expected computational time efficiency was estimated at 90 when using ARF-GATE instead of sGATE. GATE including the ARF model makes it possible to speed up GATE simulations by a factor > 100 without loss of accuracy. Simulations of patient I-131 SPECT scans become feasible in about 2 days or less using reasonable computational resources (small cluster with at least 20 CPUs).