SGAMARN: A GAN Framework for Metal Artifact Reduction in CT Imaging

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Metallic objects in CT scans cause artifacts, degrading image quality and impacting clinical diagnosis, especially in radiation therapy.
• Existing metal artifact reduction (MAR) methods struggle to balance artifact removal with anatomical accuracy.

Purpose of the Study:

• To develop a novel Generative Adversarial Network (GAN) framework for effective metal artifact reduction (MAR) in CT images.
• To enhance anatomical accuracy and image quality in CT scans affected by metallic artifacts.

Main Methods:

• A modified U-Net generator with transformer blocks and self-attention mechanisms was employed for multi-scale feature extraction.
• A conditional discriminator utilizing convolutional blocks guided artifact suppression and structural fidelity.
• The framework, named SGAMARN, was trained and evaluated using the Deeplesion synthesized dataset.

Main Results:

• The proposed SGAMARN framework demonstrated superior performance in artifact suppression and structural fidelity compared to traditional and deep learning MAR methods.
• Quantitative evaluation using SSIM and PSNR showed a 7.6% improvement in PSNR, indicating enhanced image quality.
• Ablation experiments confirmed the effectiveness of individual modules within the proposed network.

Conclusions:

• The novel GAN framework effectively reduces metal artifacts in CT images while preserving anatomical integrity.
• SGAMARN offers a robust and efficient solution for improving CT image quality in the presence of metallic objects.
• This advancement has significant implications for accurate clinical diagnosis and radiation therapy planning.