Differentiated Backprojection Domain Deep Learning for Conebeam Artifact Removal.

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Cone-beam CT (CBCT) is widely used, but artifacts increase with cone angle.
• Standard reconstruction algorithms like Feldkamp, Davis, and Kress (FDK) struggle with these artifacts.
• Iterative reconstruction methods reduce artifacts but are computationally expensive.

Purpose of the Study:

• To develop a novel deep learning approach for accurate cone-beam artifact removal.
• To overcome the limitations of existing iterative reconstruction methods in CBCT.
• To reduce runtime complexity while maintaining high-quality reconstruction.

Main Methods:

• A novel deep learning network designed on the differentiated backprojection domain.
• Data-driven inversion of an ill-posed deconvolution problem related to the Hilbert transform.
• Spectral blending technique to combine coronal and sagittal reconstruction results, minimizing spectral leakage.

Main Results:

• The proposed deep learning method effectively removes cone-beam artifacts.
• Experimental results demonstrate generalization across various conditions.
• Outperforms existing iterative methods in artifact reduction and image quality.

Conclusions:

• The novel deep learning approach offers accurate cone-beam artifact removal.
• Achieves superior performance compared to iterative methods with significantly reduced runtime.
• Presents a promising alternative for high-quality CBCT reconstruction.