Prior-image-based low-dose CT reconstruction for adaptive radiation therapy.

Area of Science:

• Medical Imaging
• Radiotherapy
• Artificial Intelligence

Background:

• Adaptive Radiotherapy (ART) requires high-quality CT images for accurate treatment planning and monitoring.
• Reducing radiation dose in ART is crucial for patient safety without compromising image quality.
• Current methods may struggle to balance dose reduction with the need for detailed anatomical information.

Purpose of the Study:

• To develop and evaluate a novel deep learning approach for low-dose CT reconstruction in ART.
• To maintain essential image quality for organ delineation and dose calculation while significantly reducing radiation exposure.
• To leverage prior CT images to improve the accuracy and robustness of reconstructions from sparse data.

Main Methods:

• Development of the Prior-aware Learned Primal-Dual Network (pLPD-UNet), a deep learning model.
• Utilizing prior CT images to enhance reconstructions from low-dose scans.
• Separate training of the network on thorax and abdomen datasets to optimize for specific anatomical regions.

Main Results:

• The pLPD-UNet achieved improved reconstruction accuracy and robustness compared to traditional methods.
• The network effectively preserved image quality crucial for precise organ delineation and dose calculation.
• Significant reduction in radiation exposure was demonstrated while maintaining diagnostic image quality.

Conclusions:

• The pLPD-UNet represents a significant advancement in ART imaging.
• This AI-driven method offers a promising approach to balance radiation safety and high-resolution imaging needs.
• The integration of prior imaging data may establish a new standard for ART quality and safety.