End-to-End Simultaneous Combined Scatter Correction and Metal Projection Segmentation for CBCT Metal Artifact Reduction Network

one-beam computed tomography (CBCT) is widely used for high-precision three-dimensional imaging in the field of nuclear science and technology (e.g., nuclear medicine radiation therapy, nondestructive testing of nuclear facilities). However, when patients with metal implants in their bodies are examined by CT, the projection data will be contaminated and the CT reconstruction will produce severe metal artifacts affecting the diagnostic accuracy. Therefore, developing high-performance metal artifact reduction (MAR) methods for CBCT is essential for clinical diagnosis. Some of the existing MAR methods ignore physical phenomena such as x-ray scattering during CT imaging. The remaining MAR methods utilize scattering correction as a preprocessing approach, and this non-end-to-end MAR method increases the complexity of the MAR process. In this article, we design an end-to-end MAR method that incorporates both scattering correction and metal segmentation in the projection domain in the same network, followed by morphological post-processing and metal artifact suppression. Both qualitative and quantitative results and analyses in simulated and real data sets illustrate the superiority of this end-to-end MAR method.