Dual-Domain Denoising Diffusion Probabilistic Model for Metal Artifact Reduction

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Image Processing

Background:

• Metal artifacts in computed tomography (CT) images hinder diagnosis and treatment planning.
• Effective metal artifact reduction (MAR) is crucial for clinical CT applications.
• Existing methods often struggle with artifact removal and image quality preservation.

Purpose of the Study:

• To develop and evaluate a novel MAR algorithm using dual-domain denoising diffusion probabilistic models (DDPM).
• To improve the quality of CT images degraded by metal artifacts.
• To address hallucination issues common in generic DDPM applications in medical imaging.

Main Methods:

• Pre-processing using linear interpolation (LI) and a convolutional neural network (CNN) for initial reprojection.
• Employing two specialized DDPM networks: one for sinogram synthesis and another for image domain optimization.
• Dual-domain approach combining sinogram and image domain processing for comprehensive artifact removal.

Main Results:

• The sinogram-domain DDPM successfully reconstructs high-quality sinograms.
• The image-domain DDPM effectively eliminates residual artifacts, significantly enhancing overall image quality.
• The proposed algorithm demonstrates superior performance compared to generic DDPMs, mitigating hallucination issues.

Conclusions:

• The dual-domain DDPM-based MAR algorithm offers a significant improvement in CT image quality.
• This method enhances the clinical applicability of DDPMs in medical imaging by overcoming limitations.
• The synergistic use of specialized DDPMs in both sinogram and image domains is key to its success.