Metal artifact reduction based on automated sinogram segmentation and adaptive multiresolution MAP reconstruction method

High density objects in the field of view (FOV) cause artifacts in medical imaging. In X-ray computed tomography (CT), there are several ways to eliminate the effects of these artifacts. This paper aims to evaluate the performance of a novel reconstruction algorithm which accurately segments the metallic regions and reconstruct sharp metal/tissue boundaries, while reducing the artifacts around the metallic regions. This algorithm uses a multilevel segmentation algorithm based on Otsu's threshold and adaptive multiresolution maximum a-posteriori expectation maximization (amMAP-EM). The qualities of Gaussian noise contaminated images were evaluated quantitatively using mean squared error and line profile analysis. The reconstructed image were compared with filtered backprojection (FBP) and maximum likelihood expectation maximization (MLEM) methods. According to the results, it is possible to reconstruct the images with more clear and sharper metal/tissue boundaries using amMAP-EM compared to MLEM and FBP, while avoiding the undesired artifacts such as blurring, streak artifacts or ringing.