Complex wavelet-based sinogram segmentation for metal artifact reduction in cone-beam CT.

Objective.Metal artifacts in cone-beam computed tomography (CBCT) arise from inconsistent projections caused by highly attenuating materials, leading to severe streaking and shading that degrade image quality and hinder clinical interpretation. This work aims to develop a robust, non-learned projection-domain (PD) method for metal artifact reduction based on analytical segmentation directly in the three-dimensional (3D) sinogram.Approach.We propose a PD metal artifact reduction method that performs metal segmentation in the 3D sinogram using the 3D dual-tree complex wavelet transform. Directional wavelet coefficients are used to extract the wavefront set and singular support associated with metal structures, followed by morphological processing to obtain a binary metal mask. The corrupted projections are then inpainted using harmonic interpolation, and the final reconstruction is obtained by combining metal-free and metal-only reconstructions.Main results.The proposed method was evaluated on both simulated and clinical CBCT datasets. It consistently achieved more accurate metal segmentation and reduced artifacts compared to conventional image-domain hard-thresholding approaches, leading to improved visual quality and fewer residual streaks. Notably, the method remains effective in challenging scenarios, including complex anatomies and cases where metal objects are partially outside the reconstruction field of view.Significance.This work demonstrates that analytically grounded, PD segmentation based on directional wavelet analysis enables effective and robust metal artifact reduction in CBCT without relying on training data. The approach offers improved interpretability and practical advantages for clinical deployment, highlighting the potential of wavefront-set-based methods for artifact reduction in tomographic imaging.