Abstract: To effectively suppress the artifacts caused by metal implants in the reconstruction process
of cone beam CT (CBCT) image, a prior-image-based metal artifact reduction method was proposed.
Firstly, the reconstructed image with metal artifacts was preprocessed by bilateral filtering, metal
threshold segmentation and tissue clustering to produce the metal image and the prior image without
metal information. Secondly, the metal image and prior image were respectively forward-projected to
produce the metal projection region and prior projection data. Then, the metal projection region was
interpolated by the prior projection data and the metal neighborhood projection data to produce the restored projection data. Finally, the CT image was reconstructed by the FDK algorithm and was
fused with the metal image to produce the final corrected image. To verify the performance of the
proposed algorithm, the metal artifact reduction experiments were carried out on the 3D Shepp-Logan
head phantom and clinical head CT data. The experimental results show that compared with the
commonly used linear-interpolation-based method and image-inpainting-based method, the corrected
image of the proposed method can keep the root-mean-square error to the minimum and the peak
signal-to-noise ratio to the maximum. This indicates that the proposed method can effectively
suppress metal artifacts while preserving image edge information.

Key words: cone beam CT, metal artifact reduction, bilateral filtering, prior image, interpolation