A hybrid approach for fast simulation of X-ray computed tomography

With the advent of powerful computers the use of Monte Carlo (MC) techniques for radiation transport simulation has become one of the most popular methods for modeling X-ray computed tomography (CT) scanners. One significant problem in the use of MC calculations is the presence of statistical uncertainties in the estimates. A simple but not practical way to decrease statistical uncertainties is to run MC simulations for sufficiently long time (large number of histories) and use efficient variance reduction techniques. In this paper, we propose a hybrid approach combining MC and analytical simulations to speed up X-ray CT modeling. In addition, an analytic method for modeling of beam hardening effect (BHE) in CT is proposed for fast generation of raw data for use in validation of BHE correction algorithms. In this method, the contribution of the primary component to the projections is calculated through analytical simulations using ray-tracing methods whereas the contribution of the scatter component is calculated using pure MC simulations. The final projections are determined by appropriate combination of both simulation results. The hybrid simulator and analytic BHE modelling were validated through comparison with experimental measurements and pure MC simulation of different phantoms performed on clinical CT scanners. In general, there is good agreement between the simulated and measured projections and reconstructed images. This work has demonstrated the feasibility of using the hybrid approach for fast and accurate modeling of CT scanners thus allowing to evaluate the effect of physical, geometrical and other design parameters on CT scanner performance and image quality. The results also validate the proposed method for analytical modeling of BHE which is a useful tool for fast generation of raw CT datasets.

I. Introduction

X-ray computed tomography (CT) scanners have inherently the tendency to produce physics-related artefacts compared to conventional planar radiography owing to the fact that the images are reconstructed from a large number of independent detector elements. The simulation of xray CT imaging to assess qualitatively and quantitatively the image formation process and interpretation and to assist the development of new detector configurations using deterministic methods and simplifying approximations have been developed mainly to improve speed of operation. Analytical x-ray CT simulators are based on projection ray-tracing methods for the three-dimensional (3-D) calculation of intersections between trajectories of photons emitted from the x-ray tube focal spot toward the detector elements and all voxels or surfaces' equations for each x-ray energy bin since the attenuation coefficients of different materials are energy-dependent [1]. Monte Carlo (MC)-based simulations are based on direct transport of photons and electrons into the materials in a 3-D geometry. One significant problem in the use of MC calculations is the presence of statistical uncertainties (noise) in the estimates [2]. A simple but not practical way to decrease statistical uncertainties is to run MC simulations for sufficiently long time (large number of histories) and use efficient variance reduction techniques. During the last decade, several research groups have investigated the issue of fast simulation of x-ray imaging [3]. It is well known that analytic algorithms based on ray-tracing techniques make it possible to simulate in a short time realistic radiographs. The analytic approach is very fast but actually limited to primary radiation modeling only (i.e., photons that do not interact in the object before being detected). On the other hand, the stochastic nature of involved processes such as x-ray photons generation, interaction with matter and detection makes MC the ideal tool for accurate modeling of x-ray imaging systems.