Deep learning‐based method for reducing residual motion effects in diffusion parameter estimation

Abstract

Conventional motion-correction techniques for diffusion MRI can introduce motion-level-dependent bias in derived metrics. To address this challenge, a deep learning-based technique was developed to minimize such residual motion effects. Methods: The data-rejection approach was adopted in which motion-corrupted data are discarded before model-fitting. A deep learning-based parameter estimation algorithm, using a hierarchical convolutional neural network (H-CNN), was combined with motion assessment and corrupted volume rejection. The method was designed to overcome the limitations of existing methods of this kind that produce parameter estimations whose quality depends strongly on a proportion of the data discarded. Evaluation experiments were conducted for the estimation of diffusion kurtosis and diffusion-tensor-derived measures at both the individual and group levels. The performance was compared with the robust approach of iteratively reweighted linear least squares (IRLLS) after motion correction with and without outlier replacement. Results: Compared with IRLLS, the H-CNN-based technique is minimally sensitive to motion effects. It was tested at severe motion levels when 70% to 90% of the data are rejected and when random motion is present. The technique had a stable performance independent of the numbers and schemes of data rejection. A further test on a data set from children with attention-deficit hyperactivity disorder shows the technique can potentially ameliorate spurious group-level difference caused by head motion. Conclusion: This method shows great potential for reducing residual motion effects in motion-corrupted diffusion-weighted-imaging data, bringing benefits that include reduced bias in derived metrics in individual scans and reduced motion-leveldependent bias in population studies employing diffusion MRI.

Effect of number of diffusion‐encoding directions in diffusion metrics of 5‐year‐olds using tract‐based spatial statistical analysis

“…As we aimed to retain as many directions as possible, the combination of DTIprep and manual exclusion was found to result in the most efficient method. In a recent study, deep learning-based method was suggested as one solution in removing motion corruption from DTI data (Gong et al, 2021). As a promising method, it also requires a training diffusion weighted dataset with similar imaging parameters to study subjects, and thus needs to be considered during study design.…”

Deep Learning for Image Enhancement and Correction in Magnetic Resonance Imaging—State-of-the-Art and Challenges

“…al. [ 49 ] • Provided estimates for both motion parameters and motion corrected images • A two-step approach where DL is used as a preprocessing step for an iterative motion correction model, and potentially multiple sources of errors may add together Ghodrati et al [ 68 ], Terpstra et al [ 75 ], Tamada et al [ 23 ] • Methods developed for dynamic MR imaging and can correct for non-rigid motion artefact including cardiac cine and DCE liver MRI • Small dataset used for training with proof of concept validation with limited clinical evaluation Khalili et al [ 83 ], Duffy et al [ 84 ], Gong et al [ 85 ], Shaw et al [ 74 ] • Focused on practical application of motion correction methods by assessing the downstream tasks such as image segmentation, cortical surface reconstruction, and diffusion parameter estimation on motion corrected images • Actual motion corrected images are not compared with ground truth images …”

“…Duffy et al [ 84 ] proposed a similar idea on T1-weighted images and used a single network with adversarial loss to enforce learning of scanner independent features. Gong et al [ 85 ] used adversarial loss to normalize data from different scanners and tested the CNN model for multiple tasks including segmentation (gray, white, CSF), regression (age prediction), and classification (gender prediction). Dewey et al [ 47 ] proposed a Unet-based model for T1, T2, FLAIR, and proton density images to modify the contrast of the source image to a predefined target image contrast, with the target image used for further processing such as segmentation.…”

Recommendation for Cardiac Magnetic Resonance Imaging-Based Phenotypic Study: Imaging Part

“…Some retrospective motion correction algorithms can be introduced to improve the image quality during the image reconstructions. Among them, several deep learning based algorithms have achieved good performances for motion correction (Duffy et al 2021 ; Gong et al 2020 ; Wang et al 2020 ). Deep learning based quality control algorithms are also useful for large cohort studies (Sujit et al 2019 ; Samani et al 2020 ; Küstner et al 2018 ).…”