A dense and U-shaped transformer with dual-domain multi-loss function for sparse-view CT reconstruction

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Sparse-view CT (SVCT) is crucial for reducing radiation dose but suffers from artifacts with traditional methods.
• Artifacts in SVCT images degrade image quality and hinder accurate diagnosis.
• Developing advanced methods to suppress these artifacts is essential for clinical applications.

Purpose of the Study:

• To develop a deep learning-based method for suppressing artifacts in sparse-view CT image reconstruction.
• To improve the quality of CT images reconstructed from sparse-view projections.

Main Methods:

• Proposed a Dense U-shaped Transformer (D-U-Transformer) architecture, combining DenseNet and Transformer.
• Leveraged densely connected convolutions for local context and Transformer for long-range dependencies.
• Incorporated channel attention and a dual-domain multi-loss function for optimization.

Main Results:

• The D-U-Transformer demonstrated superior performance in artifact suppression and feature preservation compared to other DL models on the Mayo Clinic LDCT dataset.
• Ablation studies confirmed the effectiveness of individual components within the D-U-Transformer model.
• Achieved high-precision SVCT reconstruction.

Conclusions:

• The proposed D-U-Transformer effectively suppresses sparse artifacts in CT imaging.
• This method facilitates low-dose radiation and high-quality imaging in clinical CT scanning.
• The findings are applicable to denoising and artifact removal in CT and other medical imaging modalities.