[A dual-domain cone beam computed tomography reconstruction framework with improved differentiable domain transform...

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Context:

• Cone beam computed tomography (CBCT) is crucial for medical imaging, but suffers from cone-angle artifacts.
• Existing reconstruction methods struggle with artifact correction and detail preservation.

Purpose:

• To introduce DualCBR-Net, a dual-domain CBCT reconstruction framework.
• To improve cone-angle artifact correction using a differentiable domain transform.

Summary:

• DualCBR-Net integrates projection preprocessing, differentiable domain transform (incorporating FDK and forward projection), and image post-processing.
• The framework utilizes extended projection data and geometric parameters for accurate gradient backpropagation.
• Experiments on a public chest dataset demonstrate superior artifact removal and detail preservation compared to existing methods.

Impact:

• DualCBR-Net significantly enhances CBCT image quality, improving PSNR and SSIM.
• The framework enables effective joint training of dual-domain networks for CBCT.
• It shows particular efficacy in correcting artifacts in large cone-angle regions.