Simulation of cone beam computed tomography chest imaging with parallel computing: Nodule detectability versus dose

We have developed a computer simulation model for cone beam computed tomography (CT) chest imaging on a general-purpose personal computer cluster system. Our simulation model incorporates quantum noise, detector blurring, and additive system noise. The main objective is to study how x-ray dose would affect the detectability of nodules in simulated cone beam CT chest images. The Radon transforms formalism was used to calculate the projection views for an analytically modeled chest phantom. A parallel random number generator was then used to simulate and add quantum noise whose level depends on the incident x-ray fluence, detector quantum efficiency and pixel size (0.4 mm). We also simulated detector blurring by convolving the noise added images with a Gaussian function matching the modulation transfer function measured for the flat panel x-ray detector studied. Then we modeled the additive system noise and added to the final projection images. The noise level (σ = 20) for the additive system noise was calculated from the noise power spectrum of the flat panel detector using the curve-fitting technique. The Feldkamp algorithm with a Gaussian pre-filtering process was used to reconstruct 3D image data from the projection images. For nodule contrast, the linear attenuation coefficient difference between nodule and lung was set to 10.0%. The diameters for the spherical nodules ranged from 0.2 to 1.7 cm. It was found that our Gaussian pre-filtering process helped reduce the noise level in the reconstructed images and allowed the nodules to be better visualized significantly. At 100, 000 photons per pixel (8000 mR total un-attenuated exposure at the rotating center), nodules 0.3 mm or larger could be visualized; at 10, 000 photons per pixel (800 mR), nodules 0.5 mm or larger could be visualized; at 2000 photons per pixel (160 mR), only nodules 1.5 mm or larger could be visualized.