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

Area of Science:

• Medical Imaging
• Signal Processing
• Computational Science

Background:

• Metal artifacts in cone-beam computed tomography (CBCT) severely degrade image quality due to highly attenuating materials.
• These artifacts manifest as streaking and shading, hindering accurate clinical interpretation.

Purpose of the Study:

• To develop a robust, non-learned projection-domain method for metal artifact reduction (MAR) in CBCT.
• The goal is to achieve MAR through analytical segmentation directly within the 3D sinogram.

Main Methods:

• A novel MAR method employing the 3D Dual-Tree Complex Wavelet Transform (3D DT-CWT) for metal segmentation in the 3D sinogram.
• Utilizing directional wavelet coefficients to identify metal structures, followed by morphological processing to create a binary metal mask.
• Inpainting corrupted projections with harmonic interpolation and combining metal-free and metal-only reconstructions for the final output.

Main Results:

• The proposed method demonstrated superior metal segmentation accuracy and artifact reduction compared to traditional image-domain hard-thresholding.
• Evaluations on simulated and clinical CBCT data showed improved visual quality and reduced residual streaks.
• The technique proved effective even in complex anatomies and with partially out-of-view metal objects.

Conclusions:

• Analytically grounded, projection-domain segmentation using directional wavelet analysis provides effective and robust MAR in CBCT.
• This non-learned approach enhances interpretability and offers practical advantages for clinical implementation.
• Wavefront-set-based methods show significant potential for artifact reduction in tomographic imaging.