Cross-Domain TransNet for sparse-view CT reconstruction

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Sparse-view computed tomography (CT) is vital for reducing patient radiation exposure.
• Current dual-domain reconstruction methods often process image and projection domains sequentially, missing correlations.

Purpose of the Study:

• To develop an advanced dual-domain framework for sparse-view CT reconstruction.
• To improve reconstruction quality and reduce artifacts while minimizing radiation dose.

Main Methods:

• Proposed Cross-Domain TransNet, a Transformer-based dual-domain framework.
• Integrated image and sinogram representations using a hybrid self-attention mechanism.
• Introduced a Convolution Fusion Layer (CFL) to enhance feature interactions.

Main Results:

• Demonstrated superior performance and generalization on the NIH-AAPM dataset.
• Cross-Domain TransNet improved reconstruction quality, suppressed noise, and reduced artifacts.
• Outperformed conventional algorithms and state-of-the-art deep learning methods.

Conclusions:

• Cross-Domain TransNet offers an effective and robust solution for sparse-view CT reconstruction.
• The framework fully exploits complementary information from image and projection domains.
• Enhanced diagnostic image quality and supported radiation dose reduction.