DICDNet: Deep Interpretable Convolutional Dictionary Network for Metal Artifact Reduction in CT Images.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Metallic implants in patients cause artifacts in computed tomography (CT) images, hindering clinical diagnosis and treatment.
• Existing deep learning methods for metal artifact reduction (MAR) often lack interpretability and fail to leverage intrinsic image priors.

Purpose of the Study:

• To propose a novel deep interpretable convolutional dictionary network (DICDNet) specifically designed for metal artifact reduction in CT images.
• To address the limitations of current MAR techniques by incorporating model interpretability and intrinsic image priors.

Main Methods:

• Developed a convolutional dictionary model to encode characteristic metal artifact patterns (non-local streaking, star-shape).
• Proposed a novel proximal gradient-based optimization algorithm, unfolded into interpretable network modules.
• Utilized synthesized and clinical datasets for comprehensive evaluation.

Main Results:

• The proposed DICDNet effectively reduces metal artifacts in CT images.
• DICDNet demonstrates superior interpretability compared to existing state-of-the-art MAR methods.
• Experiments validated the effectiveness on both synthesized and real-world clinical data.

Conclusions:

• DICDNet offers an effective and interpretable solution for metal artifact reduction in CT imaging.
• The approach leverages specific artifact patterns and prior knowledge for improved performance.
• This work advances MAR techniques, benefiting clinical diagnosis and patient care.