Comparison of Experimental, mantis, and hybridmantis X-ray Response for a Breast Imaging CsI Detector

Abstract

Simulations have become a very important tool in studying the details of the physical processes underlying imaging systems. With the current generation of many-core computer architectures, it has become possible to have realistic simulations in reasonable computing times. In this work, we briefly describe hybridmantis, a fast Monte Carlo tool for simulating indirect x-ray detectors that uses a hybrid approach to maximize the utilization of CPUs and GPUs in a workstation. hybridmantis is based on the mantis code with an improved geometrical model. Moreover, hybridmantis can run on GPUs for maximum computational efficiency. We compare hybridmantis results on point response and modulation transfer function for a CsI scintillator screen against experimental and mantis results. For quantitative analysis, we calculate the root mean square (RMS) difference and Swank factors for simulated and experimental data. We find that hybridmantis matches the experimental results as good as or better than mantis, especially in the high spatial frequency range. The RMS values were lower (0.025, 0.028 for 40 and 70 kVp input spectra respectively) for hybridmantis than for mantis (0.049, 0.075 respectively) when compared to experimental data. The comparison of Swank factors suggests that hybridmantis and mantis are both consistent with the experimental data. Our models of detector response are useful tools for the design and optimization of breast imaging systems and for improved description of the forward problem in reconstruction algorithms.