Metal artifact reduction in CT based on maximized the difference of mutual information segmentation

A new segmentation-based method to reduce the metal artifacts in computed tomography is proposed. The proposed method firstly uses the anisotropic Gaussian filter to suppress the noise and to smooth streak artifacts of the CT image. Next, based on maximized the difference of mutual information segmentation, the metal image is segmented from the filtered image. Then, a feedback based interpolation algorithm is used to recover the original sinogram in the projection domain. The final image is reconstructed by the filtered-back-projection method from recovered sinogram. The proposed method has been tested on real phantom and clinical CT images.