TMAA-net: tensor-domain multi-planal anti-aliasing network for sparse-view CT image reconstruction

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Deep learning methods for sparse-view CT reconstruction often use sinogram upscaling networks.
• Sinogram-based networks struggle with aliasing artifacts and low-contrast detail recovery.

Purpose of the Study:

• To introduce a view-by-view back-projection (VVBP) tensor-domain network for improved sparse-view CT image reconstruction.
• To overcome limitations of sinogram-based deep learning approaches.

Main Methods:

• Utilized a 3D tensor-domain network for direct artifact addressing, unlike 2D sinogram methods.
• Implemented multi-planar anti-aliasing modules in coronal and sagittal tensor planes.
• Incorporated a data-fidelity-based refinement module for image sharpness and texture recovery.

Main Results:

• The VVBP tensor-domain network demonstrated superior artifact removal and low-contrast detail recovery compared to state-of-the-art sinogram networks.
• Performance was validated on both numerical and clinical projection data.
• Aliasing artifacts in sparse-view CT show distinct patterns in tensor planes, enabling effective removal.

Conclusions:

• View-by-view aliasing artifacts are effectively removable in high-dimensional tensor representations.
• Tensor space processing offers better generalization for artifact removal than sinogram-domain processing.