Sinogram domain metal artifact correction of CT via deep learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Image Processing

Background:

• Metal artifacts in computed tomography (CT) images degrade quality due to X-ray attenuation by implants.
• Beam hardening artifacts, a type of metal artifact, manifest as strip artifacts, hindering clinical diagnosis and treatment planning.

Purpose of the Study:

• To propose a novel deep learning-based method for correcting beam hardening artifacts in the sinogram domain.
• To enhance the robustness of CT image reconstruction by addressing metal artifacts in specific sinogram regions.

Main Methods:

• A three-module deep learning model was developed, including a Sinogram Metal Segmentation Network (Seg-Net), a Sinogram Enhancement Network (Sino-Net), and a Fusion Module.
• The Seg-Net (Attention U-Net) identifies and segments metal regions in the sinogram.
• The Sino-Net compensates for lost information in metal regions, and a fusion module combines corrected and metal-free images for final reconstruction.

Main Results:

• The proposed method significantly improved CT image quality.
• Quantitative metrics showed substantial enhancements: Peak Signal-to-Noise Ratio (PSNR) increased from 18.22 to 30.32.
• Structural Similarity Index Measure (SSIM) improved from 0.75 to 0.99, and Weighted PSNR (WPSNR) rose from 21.69 to 35.68.

Conclusions:

• The developed deep learning approach reliably corrects beam hardening artifacts in CT images.
• The method demonstrates high accuracy in artifact correction, leading to improved diagnostic image quality.