Modified Eulers Elastica Inpainting for Metal Artifact Reduction in CT

Abstract

In computed tomography (CT) metal causes strong streak artifacts in the reconstructed images, which may hinder the correct diagnosis of patients. In this work, metal artifacts are reduced using the modified Eulers elastica and curvature based sinogram inpainting (EECSI). The inconsistent data is deleted from the sinogram and replaced by a two-dimensional interpolation strategy based on the surrounding data information using partial differential equations. The inpainting domain is defined by a binary mask that is calculated using a thresholding method in the preliminary reconstructed image with a following forward projection of the metal-only image. Here, the EECSI is modified by a statistic weighting function that regulates the time step size depending on the probability of a gray value. Additionally, the connectivity principle is strengthened by applying a different strategy in the numerical implementation compared to prior methods. The measurements are realized on a Philips Tomoscan M/EG scanner using a torso phantom with the physical properties of an average male torso, suitable for medical experiments. For evaluation, the results are compared to the ones achieved by using the original EECSI and the classically used one-dimensional linear interpolation method. It is demonstrated that the performed modifications result in a better metal artifact suppression.