SU-E-I-95: Multiscale Normalized Metal Artifact Reduction (MS-NMAR) in Computed Tomography.

PURPOSE

Medical x-ray CT devices produce images of high quality; however, the metal artifacts may arise from the adverse effect of metal materials present in the imaged objects. Conventional methods for metal artifact reduction (MAR) substitute the contaminated projection data corresponding to metal traces by specific interpolations in the projection space. Nevertheless, these methods usually introduce new artifacts, because the prior information of the imaged object (i.e., the prior image) was not involved. Therefore, a normalized MAR (NMAR) has further been invented, in which the projection space interpolation was carried out with respect to the ratio (i.e., normalization) of projection raw data to re-projection of the prior image. This feature renders NMAR a very high accuracy. However, the normalization in NMAR is more reasonable for the larger scale contents rather than for the smaller scale details.

METHODS

The proposed MS-NMAR method is a generalization of NMAR with a dedicated multiscale framework. Both projection raw data and re-projection of the prior image ill be decomposed into each scale. The normalization followed by the linear interpolation is performed in each scale, in which the larger the scale is the wider the interval the linear interpolation is operated with. Composite projection data are acquired by summing the projection components in all scales. MS-NMAR corrected image is reconstructed from the composite projection data.

RESULTS

Real CT data are used to verify the efficiency of the proposed method, and to compare with conventional MAR and NMAR. The quality of reconstructed image after MAR is evaluated by inspecting the region around the metal material.

CONCLUSIONS

Both NMAR and MS-NMAR have a better performance as compared with the conventional MAR. Meanwhile, MS-NMAR outperforms NMAR, considering it hinders the productions of new artifacts while reducing the original metal artifacts. This work is partially supported by the US National Institute of Health through grants P50-AG025688 and 2P50AG025688.