Related Experiment Video
Updated: Aug 24, 2025
05:07
Author Spotlight: Optimized Lung MRI Protocol with Computationally Efficient Reconstruction Methods
Published on: September 6, 2024
456
Undersampling artifact reduction for free-breathing 3D stack-of-radial MRI based on a deep adversarial learning
Chang Gao1, Vahid Ghodrati1, Shu-Fu Shih2
1Department of Radiological Sciences, University of California Los Angeles, Los Angeles, CA, United States; Inter-Departmental Graduate Program of Physics and Biology in Medicine, University of California Los Angeles, Los Angeles, CA, United States.
Magnetic Resonance Imaging
|October 21, 2022
Summary
A novel 3D generative adversarial network (GAN) effectively reduces streaking artifacts in undersampled stack-of-radial MRI scans, preserving crucial image sharpness for abdominal imaging.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Stack-of-radial MRI enables free-breathing abdominal scans but suffers from long acquisition times.
- Undersampling accelerates scans but introduces streaking artifacts and degrades image quality.
Purpose of the Study:
- To develop and evaluate a deep learning network for reducing streaking artifacts in undersampled stack-of-radial MRI.
- To assess the network's ability to preserve perceptual image sharpness.
Main Methods:
- A 3D generative adversarial network (GAN) was developed using adversarial, mean-squared-error, and structural similarity index loss functions.
- Training data was augmented by gating to five respiratory states to mitigate motion artifacts.
- The GAN's performance was evaluated on 3-5x undersampled data from two institutions, comparing it against a 3D U-Net.
Main Results:
- The 3D GAN achieved significantly higher SSIM (0.841 vs. 0.798) compared to the 3D U-Net.
- Region of interest analysis confirmed effective streak removal without altering tissue characteristics.
- Radiologist scores indicated a significant improvement in artifact reduction (1.6 points on a 4-point scale) with no compromise in sharpness.
Conclusions:
- The developed 3D GAN effectively removes streaking artifacts from undersampled stack-of-radial abdominal MRI.
- This deep learning approach successfully preserves perceptual image details, enhancing diagnostic quality.
