A deep learning reconstruction framework for X-ray computed tomography with incomplete data

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Science

Background:

• X-ray computed tomography (CT) is a vital non-destructive imaging technique for internal structure analysis.
• Incomplete projection data in CT, common due to practical constraints, poses significant challenges for conventional reconstruction algorithms.
• Existing methods often require complex parameter tuning, are susceptible to noise, and are computationally intensive.

Purpose of the Study:

• To develop a deep learning-based reconstruction framework capable of handling incomplete projection data in CT.
• To improve the quality and efficiency of CT image reconstruction, particularly for sparse-view and limited-angle scenarios.

Main Methods:

• A deep learning framework integrating a U-Net architecture with CT reconstruction algorithms was developed.
• The U-Net was trained to estimate complete projection sinograms from incomplete data, rather than directly reconstructing images.
• The framework was validated using both synthetic and experimental datasets for sparse-view and limited-angle CT.

Main Results:

• The proposed deep learning framework successfully reconstructed high-quality CT images from incomplete projection sinograms.
• The method effectively addressed artifacts typically associated with incomplete data reconstruction.
• Validation demonstrated robust performance in challenging sparse-view and limited-angle CT scenarios.

Conclusions:

• The integrated deep learning framework offers a powerful and efficient solution for CT reconstruction with incomplete data.
• This approach overcomes limitations of traditional algorithms, offering improved image quality and reduced complexity.
• The framework's design facilitates extension to other CT imaging challenges, advancing deep learning applications in medical imaging.