Magnetic resonance-based computed tomography metal artifact reduction using Bayesian modelling.

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Computational Biology

Background:

• Metal artifacts in CT scans complicate radiotherapy planning.
• Magnetic Resonance (MR) imaging offers complementary anatomical information.
• Integrating MR data for metal artifact reduction (MAR) faces challenges due to intensity ambiguity and registration errors.

Purpose of the Study:

• To develop and evaluate a self-tuning Bayesian model for MR-based MAR.
• To address the challenges of integrating MR and CT data for artifact correction.
• To improve image quality and quantitative accuracy in CT images with metal implants.

Main Methods:

• A self-tuning Bayesian model was proposed, integrating local MR image intensities with initial CT data.
• The model was applied to three MAR scenarios: image inpainting, sinogram inpainting, and iterative reconstruction.
• A retrospective study on nine head-and-neck patients with CT and MR scans was conducted.

Main Results:

• The proposed model demonstrated improvements in image quality and quantitative CT value accuracy across all tested MAR scenarios.
• Significant enhancements were observed compared to conventional MAR methods.
• The model effectively leveraged anatomical information from co-acquired MR scans.

Conclusions:

• The self-tuning Bayesian model offers a versatile approach for MR-based MAR.
• This method enhances the performance of MAR algorithms by utilizing MR image information.
• The findings suggest a promising direction for improving CT image accuracy in the presence of metal artifacts.