Self-supervised Projection Denoising for Low-Dose Cone-Beam CT.

Area of Science:

• Medical Imaging
• Computer Vision
• Image Processing

Background:

• Low-dose X-ray imaging in cone-beam CT (CBCT) is crucial for reducing radiation exposure.
• Poisson noise significantly degrades the quality of low-dose projection data.
• Accurate reconstruction of 3D volumes from noisy projections remains a challenge.

Purpose of the Study:

• To develop a self-supervised deep learning method for denoising low-dose X-ray projection data in CBCT.
• To improve the quality of reconstructed 3D volumes by reducing noise and preserving structural information.
• To enable high-quality CBCT reconstruction without requiring ground truth data.

Main Methods:

• A convolutional neural network (CNN) was trained using a self-supervised approach.
• The model was trained on partially blinded noisy projection views to recover missing information.
• Denoised projection views were then used with filtered backprojection for 3D volume reconstruction.

Main Results:

• The self-supervised denoising method significantly reduced noise levels in low-dose X-ray projections.
• Structural details within the projection views were effectively restored.
• Reconstructed 3D volumes showed improved quality and clarity compared to standard methods.

Conclusions:

• Self-supervised learning offers a viable solution for denoising low-dose X-ray projections in CBCT.
• The proposed method enhances image quality and structural fidelity in 3D reconstructions.
• This approach holds promise for safer and more effective CBCT imaging.