Metal artifact reduction in CT based on adaptive steering filter and nonlocal sinogram inpainting

This paper proposed a novel metal artifact reduction strategy for computerized tomography (CT). Firstly, an adaptive steering filter is applied to suppress noise and smooth artifacts in the original CT image with metal artifacts. Secondly, the metal component and artifact component is extracted from the filtered image using Mutual Information Maximized Segmentation (MIMS) with different maximum class number. With tagged projection sinogram from the metal artifact part in original CT image, we complete the subtracted sinogram using a novel nonlocal means inpainting method. Then, a corrected image is obtained using the filtered back-projection of the completed sinogram. The final image is formed from a combination of the corrected image and the image component free of artifacts. Experiment with both phantom and clinical images show that CT metal artifacts can be significantly reduced using our proposed method.