Model Image-Based Metal Artifact Reduction for Computed Tomography

Metal implants often produce severe artifacts in the reconstructed computed tomography (CT) images, causing information and image detail loss and making the CT images diagnostically unusable. In order to eliminate the metal artifacts and enhance the diagnostic value of the reconstructed CT images, a post-processing metal artifact reduction algorithm, based on a tissue-class model segmented by thresholding and k-means clustering with spatial information, is proposed. The image inpainting technique is incorporated into the algorithm to improve the segmentation accuracy for CT images severely corrupted by metal artifacts. A study of a water phantom and of two sets of clinical CT images was performed to test the algorithm performance. The proposed method effectively eliminates typical metal artifacts, restores the average CT numbers of different tissues to the proper levels, and preserves the edge and contrast information, thus allowing the accurate reconstruction of the tissue attenuation map. The quality of the artifact-corrected CT images allows them to be subsequently used in other clinical applications, such as three-dimensional rendering, dose estimation for radiotherapy, attenuation correction for PET and SPECT, etc. The algorithm does not rely on the use of the raw sinogram and so is not limited by the proprietary format restrictions.