A statistical X-ray upsampling technique for enhanced spatial resolution in computed tomography

Abstract

To recover a high-resolution sinogram from the original X-ray measurements, we incorporate a physics model into the upsampling process. Since the number of the pencil beams of the upsampled geometry is greater than the number of the original measurements, the information on the geometry and the statistics of the X-ray data are essential factors for a more accurate estimate. To incorporate X-ray geometric information, we propose an iterative algorithm based on a statistical learning approach comprising three steps: ray-splitting, image reconstruction, and reprojection steps. In the ray-splitting step, we estimate the high-resolution sinogram by maximizing a posteriori of the high-resolution sinogram with the original measurements and CT model. In the image reconstruction step, we reconstruct the high-resolution CT image with the estimated high-resolution sinogram. In the reprojection step, we update the statistical model of the high-resolution sinogram by reprojecting the reconstructed image for the next ray-splitting step. The performance of the proposed algorithm is demonstrated through a series of phantom experiments to evaluate the enhancement of spatial resolution in the reconstructed CT images. The results show that the statistically recovered high-resolution sinogram successfully improves spatial resolution without hardware changes.