Modified MLEM Algorithm for Artifact Suppression in CT

Iterative methods for CT image reconstruction are widely disregarded in clinical routine due to the inherent massive computational effort. On the other hand, the continuously growing computational power of today's standard computers has led to a rediscovery of these methods. Currently, filtered backprojection (FBP) is the reconstruction method of choice, because it is very fast, especially on dedicated hardware. However, a disadvantage of FBP is that it ignores some of the physical effects like beam hardening, scatter and noise, leading to artifacts in the reconstructed images. Artifacts of this type are particularly dominant if metal objects are inside the patient, because FBP interprets the corresponding projection data as inconsistent. For the method proposed here, the inconsistency of the metal-influenced projection data is reduced with an interpolation scheme working in the Radon-space domain. As the interpolation yields a vague estimate of the real soft-tissue projection values, residual inconsistencies remain. Those residuals can be treated by a modified maximum likelihood expectation maximization (MLEM) method presented in this paper. Within the novel approach, different projection lines through the object can be associated with appropriate weightings that decrease the influence of the residual inconsistencies - resulting in an overall image-quality enhancement. The approach is demonstrated on real CT measurements of a torso phantom equipped with metal markers that allows an evaluation with the corresponding ground truth data.