I. INTRODUCTION
When metal objects are present in x-ray computed tomography (CT) scans, they are accompanied by bright and dark shadows and streaks, collectively called metal artifacts. These artifacts are due to physical processes that cause the assumption of linearity in the reconstruction to break down, so that the scanner projections cannot be accurately reconstructed using filtered backprojection or Radon transform inversion. The artifacts obscure information about anatomical structures, making it difficult for radiologists to correctly interpret the images or for computer programs to analyze them. The problem has existed for many years,1,2 and there has been re- cent progress, but comparisons across different methods have not been made other than with linear interpolation across the metal traces.1–13 As a result, there is no robust or widely accepted solution, and it continues to be a challenging research problem.
The separation of artifact from real tissue is not a trivial task. The CT number (voxel intensity) ranges of metal artifacts and anatomical structures overlap, as do their gradient ranges. Metal artifact reduction (MAR) algorithms operate in projection space, so that the scanner projection samples, i. e., measured ray-sums, that do not pass through metal contribute to the final image, along with estimated ray-sums that replace the measured ray-sums passing through the metal. In this
5857 Med. Phys. 39 (10), October 2012 0094-2405/2012/39(10)/5857/12/$30.00 © 2012 Am. Assoc. Phys. Med. 5857