GDP-Net: Global Dependency-Enhanced Dual-Domain Parallel Network for Ring Artifact Removal

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Ring artifacts in Computed Tomography (CT) imaging degrade image quality due to inconsistent detector responses, particularly severe in photon-counting detector systems.
• Modeling ring artifacts is challenging due to the diverse detector responses and their global nature across coordinate systems.

Purpose of the Study:

• To propose a novel deep learning approach for effective Ring Artifact Removal (RAR) in CT images.
• To enhance the performance of artifact removal by exploiting features in both Cartesian and Polar domains.
• To address the limitations of Convolutional Neural Networks in modeling long-range dependencies inherent in ring artifacts.

Main Methods:

• A dual-domain parallel neural network architecture processing images in both Cartesian and Polar coordinates.
• Integration of the Mamba mechanism to efficiently capture global dependencies and long-range relationships within the artifact structure.
• Validation using simulated data and evaluation on two unseen real-world CT datasets.

Main Results:

• The proposed dual-domain parallel network effectively extracts features from different coordinate systems for improved RAR.
• The Mamba mechanism significantly enhances the model's ability to capture global dependencies, crucial for artifact reduction.
• Experimental results demonstrate superior performance in eliminating ring artifacts and preserving image details compared to existing methods.

Conclusions:

• The developed global dependency-enhanced dual-domain parallel neural network offers a promising solution for severe ring artifact issues in CT imaging.
• The Mamba mechanism integration is key to achieving high-performance artifact removal by modeling long-range dependencies effectively.
• This approach shows significant potential for improving the diagnostic accuracy and utility of CT imaging, especially with advanced detector technologies.