A dual-domain neural network based on sinogram synthesis for sparse-view CT reconstruction.

Area of Science:

• Medical Imaging
• Deep Learning
• Image Reconstruction

Background:

• Dual-domain deep learning methods are used for medical image reconstruction.
• Sparse-view reconstruction in computed tomography (CT) can lead to streak artifacts due to missing data.
• Existing dual-domain techniques struggle to adequately suppress artifacts in sparse-view scenarios.

Purpose of the Study:

• To propose an intelligent sinogram synthesis based back-projection network (iSSBP-Net) for sparse-view CT reconstruction.
• To overcome the limitations of insufficient artifact suppression in dual-domain methods for sparse-view CT.
• To enhance the quality of CT images reconstructed from limited projection data.

Main Methods:

• Developed an end-to-end network, iSSBP-Net, integrating a sinogram synthesis sub-network (SS-Net), a sinogram filter sub-network (SF-Net), a back-projection layer, and a post-CNN.
• Utilized CNN within SS-Net to synthesize full-view sinograms by inpainting missing data.
• Filtered synthesized sinograms with SF-Net to improve visual quality before back-projection and final image restoration.

Main Results:

• The iSSBP-Net demonstrated superior performance compared to existing algorithms in sparse-view CT reconstruction across various view conditions.
• Achieved improvements in peak signal-to-noise ratio (PSNR) of 1.21 dB (360 views), 0.26 dB (180 views), 0.01 dB (90 views), and 0.37 dB (60 views).
• Numerical experiments confirmed the effectiveness of the proposed method under different scanning conditions.

Conclusions:

• Integrating a sinogram synthesis network (SS-Net) into dual-domain methods is an effective strategy for suppressing streak artifacts in sparse-view CT images.
• The iSSBP-Net method shows promising results for sparse-view CT reconstruction.
• This study encourages further research into dual-domain methods incorporating sinogram synthesis for improved CT image reconstruction.