GPU-based fast cone beam CT reconstruction from undersampled and noisy projection data via total variation

Area of Science:

• Medical Physics
• Image Reconstruction
• Computational Imaging

Background:

• Cone-beam CT (CBCT) is crucial for image-guided radiation therapy (IGRT).
• High radiation doses from serial CBCT scans limit its use, particularly in pediatric patients.
• Developing low-dose CBCT reconstruction methods is essential for broader clinical application.

Discussion:

• A novel GPU-based algorithm reconstructs CBCT from undersampled and noisy projection data.
• The algorithm minimizes an energy functional using data fidelity and total variation regularization.
• A multigrid technique and a GPU-friendly forward-backward splitting method accelerate reconstruction.

Key Insights:

• Satisfactory CBCT image quality for IGRT can be achieved with as few as 20-40 X-ray projections.
• Reconstruction times are significantly reduced (77-130 seconds) on a GPU, approximately 100x faster than traditional iterative methods.
• Phantom studies demonstrate a 36-72 fold reduction in radiation dose compared to standard protocols.

Outlook:

• The developed algorithm enables substantial reduction in imaging dose for CBCT.
• High computational efficiency makes iterative CBCT reconstruction feasible in clinical settings.
• This advancement holds promise for expanding IGRT accessibility, especially for dose-sensitive populations.