Multi-stage deep learning framework reduces truncation and metal artifacts in half-detector CBCT

A multi-stage deep learning framework (HD-TMAR) suppressed combined truncation and metal artifacts in half-detector CBCT, achieving the highest fidelity and preserving dental morphology in simulations.

• HD-TMAR uses a three-stage strategy: sinogram correction to isolate artifact residuals, specialized overlapping patching for reconstruction, and image refinement for anatomical textures.
• In realistic simulations, HD-TMAR outperformed previous deep learning metal artifact reduction methods qualitatively and quantitatively, with no secondary artifacts or blurring.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.