Prior-based metal artifact reduction in CT using statistical metal segmentation on projection images

Metal objects such as dental fillings, dental implants, and surgical clips generate metal artifacts appearing as streaks and dark shadows. We present a method that can reduce metal artifacts by generating a prior image. The prior image was used to replace the pixel values at the metal region in the projection image. To generate the prior image, we first segmented the metal parts in the projection image using simple global thresholding. We corrected the result of simple global thresholding by applying local thresholding based on statistics of pixel values around the metal boundary. Then, we applied linear interpolation to initially fill the metal region with pixel values similar to the surrounding body tissues. From the interpolated projection data, we reconstructed the first image for generating the prior image. We reconstructed the second image from the original projection data with removing the metal parts. We generated the prior image by weighted summation of the two images. Then, we synthesized a projection data by forward projecting the prior image. For the final image reconstruction, we replaced the metal parts in the original projection data with the ones in the synthesized projection data. The proposed method shows significantly improved performance in reducing streak artifacts and dark shadows in experimental imaging studies of a dental phantom using a micro-CT.