Image-based motion artifact reduction on liver dynamic contrast enhanced MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Motion artifacts significantly degrade the quality of liver Magnetic Resonance Imaging (MRI), impacting diagnostic accuracy.
• Respiratory and bulk motion during scans introduce ghosting and blurring, necessitating artifact reduction techniques.

Purpose of the Study:

• To develop and evaluate a novel two-stage deep learning model for motion artifact reduction in dynamic contrast-enhanced (DCE) liver MRIs.
• To preserve anatomical details and image quality in liver MRIs affected by motion.

Main Methods:

• A two-stage deep learning approach was employed, utilizing a deep residual network with a densely connected multi-resolution block (DRN-DCMB) for initial artifact removal.
• A generative adversarial network (GAN) with perceptual loss compensation was used in the second stage to refine image quality and preserve structural features.
• The model was trained and validated using simulated and real motion artifacts on liver DCE-MRI datasets.

Main Results:

• The two-stage model successfully reduced motion artifacts in liver MRIs.
• Quantitative evaluation showed high performance with SSIM of 0.935, MSE of 60.7 × 10⁻³, and PSNR of 32.054.
• The processed images retained anatomical details without introducing blurriness.

Conclusions:

• The proposed two-stage deep learning model is effective in mitigating motion artifacts in liver DCE-MRIs.
• This approach offers a promising solution for improving the diagnostic quality of liver MRI examinations.
• The method preserves essential image features, ensuring diagnostic utility.