Motion artefact reduction in coronary CT angiography images with a deep learning method

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• Motion artifacts are a common problem in coronary CT angiography (CCTA).
• These artifacts can obscure important diagnostic information in cardiac scans.
• Developing methods to reduce motion artifacts is crucial for accurate CCTA interpretation.

Purpose of the Study:

• To evaluate the effectiveness of a pixel-to-pixel generative adversarial network (GAN) in removing motion artifacts from CCTA images.
• To assess the quantitative and qualitative improvements in CCTA image quality after GAN-based artifact reduction.

Main Methods:

• A pixel-to-pixel GAN was trained using raw CCTA images and SnapShot Freeze (SSF) CCTA images from 97 patients.
• The GAN learned to generate artifact-reduced CCTA images from raw data.
• Image quality was assessed using structural similarity (SSIM), Dice similarity coefficient (DSC), circularity index, and visual assessment by radiologists.

Main Results:

• GAN-generated images showed significantly higher circularity (0.82 vs. 0.74) and SSIM (0.87 vs. 0.83) compared to raw CCTA images.
• The DSC indicated a significantly higher overlap between GAN-generated and SSF images than between GAN-generated and raw images.
• Motion artifact scores were significantly reduced in GAN-generated images of the RCA.

Conclusions:

• A GAN can significantly reduce motion artifacts in coronary CT angiography images.
• This AI-driven approach shows potential as a novel method for improving CCTA image quality.
• Further research may establish GANs as a standard tool for artifact removal in cardiac imaging.