Related Experiment Video
Updated: Jan 1, 2026
06:06
Optogenetic Functional MRI
Published on: April 19, 2016
15.3K
Suppressing motion artefacts in MRI using an Inception-ResNet network with motion simulation augmentation
Kamlesh Pawar1,2, Zhaolin Chen1, N Jon Shah1,3
1Monash Biomedical Imaging, Monash University, Melbourne, Australia.
NMR in Biomedicine
|December 23, 2019
Summary
This study introduces a new deep learning method to remove motion artifacts in MRI scans. The technique effectively suppresses artifacts, improving image quality for clinical use.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Image Processing
Background:
- Motion artifacts significantly degrade Magnetic Resonance Imaging (MRI) quality.
- Developing effective methods for motion artifact suppression is crucial for accurate diagnosis.
Purpose of the Study:
- To develop a novel, standalone deep learning technique for suppressing motion artifacts in MRI.
- To create a data-driven approach that can be applied post-acquisition.
Main Methods:
- A simulation framework generated motion-corrupted MR images.
- An Inception-ResNet encoder-decoder network was trained on simulated data.
- The network was validated on both simulated and real-world in vivo brain MRI datasets.
Main Results:
- The deep learning model successfully suppressed motion artifacts in simulated and real-world MRI.
- Mean Structural Similarity Index (SSIM) improved from 0.9058 to 0.9338 (simulated) and 0.8671 to 0.9145 (real-world).
- The method outperformed iterative entropy minimization, showing 5-10% better performance in SSIM and normalized root mean squared error.
Conclusions:
- A novel, data-driven technique effectively suppresses motion artifacts in MRI.
- The standalone, post-processing method is suitable for routine clinical practice without altering acquisition parameters.
