Sparsity driven metal part reconstruction for artifact removal in dental CT

Abstract

Metal artifact removal (MAR) is one of the most important issues in x-ray CT reconstruction. Various methods have been suggested for metal artifact removal, among which projection modification and iterative methods are most popular. While those methods mainly focus on removing background artifacts, for some applications such as dental CT the correct reconstruction of metallic inserts is also important. For this application, we formulate the MAR problem as a sparse recovery problem since metallic inserts usually occupy very little volume within a field of view. One of the main advantages of this approach is to overcome the inconsistency of sinograms from metal artifacts by imposing a geometric constraint, "sparsity". As a side product of this formulation, a significant reduction of the sample views is feasible for metal part reconstruction without sacrificing quality, thanks to the compressed sensing theory, which minimizes the additional computational overhead. Numerical results confirm that metallic inserts can be accurately reconstructed with a significant reduction of computation time.