Penalized-likelihood sinogram smoothing for dose reduction in computed tomography

摘要

We have developed a statistically principled sinogram smoothing approach for single-slice, multi-slice, and conebeam computed tomography (CT) with the aim of obtaining high-quality reconstructed images from scans conducted with lower radiation doses than are generally employed. Reducing patient dose in x-ray computed tomography (CT) can be achieved by reducing the output of the x-ray tube but this will generally increase image noise levels as well. Standard approaches to noise control in CT involve simple apodization of the reconstruction filter, which would unacceptably compromise resolution to achieve the noise control necessary in low-dose situations. In this work, we present an explicitly statistical approach to sinogram smoothing in which we formulate the estimation of the attenuation line integrals from the measured data as a statistical estimation problem and make use of penalized likelihood methods to perform the estimation. We find that the proposed penalized-likelihood sinogram smoothing approach can reduce the appearance of noise in reconstructed images without introducing additional artifacts or severely degrading resolution. In terms of resolution-noise tradeoffs, it was found to outperform other approaches considered.