Unsupervised MR Motion Artifact Deep Learning using Outlier-Rejecting Bootstrap Aggregation

Gyutaek Oh,Jeong Eun Lee,Jong Chul Ye

DOI: https://doi.org/10.48550/arXiv.2011.06337

2020-11-12

Abstract:Recently, deep learning approaches for MR motion artifact correction have been extensively studied. Although these approaches have shown high performance and reduced computational complexity compared to classical methods, most of them require supervised training using paired artifact-free and artifact-corrupted images, which may prohibit its use in many important clinical applications. For example, transient severe motion (TSM) due to acute transient dyspnea in Gd-EOB-DTPA-enhanced MR is difficult to control and model for paired data generation. To address this issue, here we propose a novel unsupervised deep learning scheme through outlier-rejecting bootstrap subsampling and aggregation. This is inspired by the observation that motions usually cause sparse k-space outliers in the phase encoding direction, so k-space subsampling along the phase encoding direction can remove some outliers and the aggregation step can further improve the results from the reconstruction network. Our method does not require any paired data because the training step only requires artifact-free images. Furthermore, to address the smoothing from potential bias to the artifact-free images, the network is trained in an unsupervised manner using optimal transport driven cycleGAN. We verify that our method can be applied for artifact correction from simulated motion as well as real motion from TSM successfully, outperforming existing state-of-the-art deep learning methods.

Computer Vision and Pattern Recognition,Machine Learning,Image and Video Processing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the motion artifact problem in magnetic resonance imaging (MRI), especially for those complex motion artifacts that are difficult to be processed by traditional methods or existing deep - learning methods. Specifically, the article focuses on the motion artifacts caused by transient severe motion (TSM) due to acute transient dyspnea in gadoxetic acid disodium (Gd - EOB - DTPA) - enhanced liver MRI.

### Main problems
1. **Lack of paired data**: Most existing deep - learning methods require paired artifact - free and artifact - containing images for supervised training, which is difficult to achieve in actual clinical applications, especially in the case of TSM.
2. **Complex motion artifacts**: Complex motion artifacts such as TSM are difficult to simulate and control, making traditional methods based on simulated data less effective.
3. **Computational complexity and parameter adjustment**: Some existing methods, such as the compressed sensing (CS) algorithm, are effective but have high computational complexity and difficult hyper - parameter adjustment.

### Solutions
To solve these problems, the author proposes a new unsupervised deep - learning framework, using outlier - rejecting bootstrap aggregation. The main innovations of this method include:
- **No need for paired data**: The training process only requires artifact - free images and does not need the corresponding artifact - containing images.
- **Outlier - rejection mechanism**: By observing that motion artifacts usually cause sparse k - space outliers in the phase - encoding direction, k - space subsampling is used to remove these outliers.
- **Bootstrap aggregation**: By multiple random subsampling and aggregating the results, the reconstruction quality is further improved.
- **Cycle GAN**: An optimal - transport - driven cycle GAN is used for unsupervised training to avoid potential biases.

### Method advantages
- **Flexibility**: It can be applied to various types of motion artifacts, including simulated data and real data.
- **Efficiency**: Compared with traditional compressed - sensing methods, it has lower computational complexity and is easier to implement.
- **Robustness**: It performs well in dealing with complex motion artifacts (such as TSM), and can effectively remove artifacts and maintain image details.

Through this method, the author has successfully solved the challenges encountered by existing methods in dealing with complex motion artifacts and demonstrated its superior performance on simulated and real data.