Artifact removal using a hybrid-domain convolutional neural network for limited-angle computed tomography imaging

Area of Science:

• Medical Imaging
• Computer Vision
• Image Processing

Background:

• Streak artifacts are a significant challenge in limited-angle computed tomography (CT).
• Conventional methods like filtered backprojection (FBP) and iterative algorithms have limitations in artifact suppression, especially for large streaks.
• Existing deep learning approaches process data in separate domains, failing to leverage combined strengths.

Purpose of the Study:

• To develop and validate a novel hybrid-domain convolutional neural network (hdNet) for effective streak artifact reduction in limited-angle CT.
• To address the limitations of current methods by integrating processing across different data domains.

Main Methods:

• A hybrid-domain convolutional neural network (hdNet) was designed, incorporating three components: a sinogram-domain CNN, a domain transformation operation, and a CT image-domain CNN.
• The network was trained to directly generate artifact-suppressed CT images from sinogram data.

Main Results:

• The proposed hdNet demonstrated significant reduction of serious streak artifacts.
• Verification using numerical, experimental, and clinical data confirmed the method's efficacy.
• The hybrid approach successfully combined the advantages of deep learning in different domains.

Conclusions:

• The hdNet offers a powerful and effective solution for streak artifact suppression in limited-angle CT.
• This hybrid-domain approach represents a significant advancement over existing methods.
• The method shows promise for improving the quality of CT images in challenging acquisition scenarios.