Metal Artifact Reduction in Computed Tomography Images by Variational Inpainting Methods
Tuesday 9th April 2013
Faggiano, E. ; Lorenzi, T. ; Quarteroni, A.
Permanent metallic implants such as dental fillings, hip prostheses and cardiac devices generate streaks-like artifacts in computed tomography images. In this paper, two methods based on partial differential equations (PDEs), the Cahn-Hilliard equation and the TV-H-1 inpainting equation, are proposed to reduce metal artifacts. Although already profitably employed in other branches of image processing, these two fourth-order variational methods have never been used to perform metal artifact reduction. A systematic evaluation of the performances of the two methods is carried out. Comparisons are made with the results obtained with classical linear interpolation and two other PDE-based approaches using, respectively, the Fourier heat equation and a nonlinear version of the heat equation relying on total variation flow. Visual inspection of both synthetic and real computed tomography images, as well as computation of similarity indexes, suggest that the Cahn-Hilliard method behaves comparably with more classical approaches, whereas the TV-H-1 method outperforms the others as it provides best image restoration, highest similarity indexes and for being the only one able to recover hidden structures, a task of primary importance in the medical field.