PAINT: Prior-Aided Alternate Iterative NeTwork for Ultra-Low-Dose CT Imaging Using Diffusion Model-Restored Sinogram.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Repeated computed tomography (CT) scans increase patient radiation dose and cancer risk.
• Clinical needs like lung nodule screening and image-guided radiation therapy often require multiple CT scans.

Purpose of the Study:

• To develop an ultra-low-dose CT imaging technique for subsequent scans.
• To preserve image quality by using preceding full-sampled scans as prior information.
• To reduce radiation exposure for patients undergoing repeated CT scans.

Main Methods:

• Proposed a two-stage framework: diffusion model-based sinogram restoration and deep learning-based unrolled iterative reconstruction.
• Utilized regional few-view scanning to collect extremely undersampled sinogram data.
• Developed Prior-aided Alternate Iterative NeTwork (PAINT) for optimization, integrating sinogram and image-domain priors.

Main Results:

• Achieved superior CT value accuracy and image detail preservation in simulated data (112 mm field of view).
• Demonstrated significant artifact reduction and structure recovery in clinical data experiments.
• Outperformed comparison methods in both simulated and clinical datasets.

Conclusions:

• The proposed framework enables ultra-low-dose CT imaging with high image quality.
• This method effectively reduces radiation exposure in scenarios requiring repeated CT scans.
• PAINT framework shows promise for safer and more effective medical imaging.