Abstract ID: 195 A fast Cone-Beam CT scatter correction method with Monte Carlo simulation using GPUMCD

Abstract

GPUMCD, a fast GPU-based Monte Carlo transport code, was initially developed and validated for photon and electron dose calculations in radiation oncology. In this work, the code was adapted to rapidly simulate realistic X-ray Cone-Beam Computed Tomography (CBCT) projections, and in particular to estimate the scatter component of the signal. This estimate is thereafter filtered and used for scatter correction, which is especially important in a CBCT geometry. The signal generated by scattered photons on a 1028 × 768 detector was estimated with GPUMCD. The simulation geometry was derived from actual CBCT reconstructions (pelvis, thorax), with density and materials assigned to each voxel with a HU-electronic density calibration curve. Simulations were conducted on a NVIDIA GTX970m GPU with either 10 7 or 10 9 photons per projection. Recursive Gaussian filtering in projections with 10 7 photons was performed to assess the potential of faster simulations conducted with fewer photons. A simple Gaussian filter was used for the projection of 10 9 photons. The primary signal was obtained with deterministic ray-tracing, and normalized by the primary signal computed derived from the Monte Carlo simulation. Simulated projections with 10 7 photons were obtained in 3.5 s while the primary signal estimation took 0.2s. Recursive Gaussian filtering of projections was shown to be an effective way to reduce the simulation time by a factor 100 and preserve more complex scatter patterns. The simulation times reported here let envision a Monte Carlo-based scatter correction strategies in CBCT reconstruction.