TIME-Net: Transformer-Integrated Multi-Encoder Network for limited-angle artifact removal in dual-energy CBCT

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computed Tomography

Background:

• Dual-energy cone-beam computed tomography (DE-CBCT) offers material-specific information but faces challenges with artifacts.
• Information redundancy in DE-CBCT data can be leveraged to improve image quality.

Purpose of the Study:

• To develop an advanced deep learning model, TIME-Net, for artifact reduction in DE-CBCT.
• To enhance image quality and explore clinical applications of DE-CBCT.

Main Methods:

• Development of the Transformer-Integrated Multi-Encoder Network (TIME-Net) with image, prior, and transformer encoders.
• Utilizing a feature fusion module and differential learning decoders for image restoration.
• Employing complementary quarter-scans for reconstruction.

Main Results:

• TIME-Net successfully removed limited-angle artifacts from DE-CBCT images.
• Demonstrated preservation of subtle structures and reconstruction accuracy in simulated and real data.
• Validated potential for virtual non-contrast imaging and iodine quantification.

Conclusions:

• TIME-Net significantly improves DE-CBCT image quality, offering artifact removal and enhanced detail.
• The technique holds potential for reducing radiation dose and acquisition time.
• TIME-Net shows promise for clinical applications like VNC imaging and iodine quantification.