Authors

Jiazhen Wang, Yizhe Yang, Yan Yang, Jian Sun

Abstract

Magnetic resonance imaging (MRI) may degrade with motion artifacts in the reconstructed MR images due to the long acquisition time. In this paper, we propose a dual domain motion correction network (D2MC-Net) to correct the motion artifacts in 2D multi-slice MRI. Instead of explicitly estimating the motion parameters, we model the motion corruption by k-space uncertainty to guide the MRI reconstruction in an unfolded deep reconstruction network. Specifically, we model the motion correction task as a dual domain regularized model with an uncertainty-guided data consistency term. Inspired by its alternating iterative optimization algorithm, the D2MC-Net is composed of multiple stages, and each stage consists of a k-space uncertainty module (KUModule) and a dual domain reconstruction module (DDR-Module). The KU-Module quantifies the uncertainty of k-space corruption by motion. The DDR-Module reconstructs motion-free k-space data and MR image in both k-space and image domain, under the guidance of the k-space uncertainty. Extensive experiments on fastMRI dataset demonstrate that the proposed D2MC-Net outperforms state-of-the-art methods under different motion trajectories and motion severities.

Link to paper

DOI: https://doi.org/10.1007/978-3-031-43999-5_28

SharedIt: https://rdcu.be/dnwwH

Link to the code repository

https://github.com/Jiazhen-Wang/D2MC-Net-main

Link to the dataset(s)

N/A

Reviews

Review #2

Please describe the contribution of the paper

The paper suggests a dual domain reconstruction framework that is guided by k-space uncertainty for motion correction of MRI images. Image- and k-space domain reconstruction blocks are conventional unet, followed by an uncertainty data consistency module.
Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

The submission is clearly written, and extensive experiments (quantitative ablation study, hyperparameter optimization, different motion levels in both image and k-space domains, etc) were run for the simulated analysis.
Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
- Computational cost and run time were not reported/compared between different methods.
- The effect of motion is not only displacing image voxels, and intensity inhomogeneity/blurring would also occur as a result of it.however, these effects were neglected in the simulation. For example, your forward model for simulating motion-corrupted images is not correct and lacks a “blurring” operator (= Gaussian psf).
- why did you select simulated motion vectors from Gaussian distribution rather than the uniform distribution?
- introduction can be improved and motion correction methods using k-space information better to be added (interlacer, autofocusing, etc)
Please rate the clarity and organization of this paper

Very Good
Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance

The paper meets the reproducibility criteria. data is from a publicly available dataset, and the reviewer didn’t find commitment by the authors to share the code/traing model upon accaptance.
Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html
- Simulated intensity inhomogeneity artifacts are also usually added to the simulated motion-corrupted slice (and then Fourier transform is applied).
- It would be interesting to test your model on real data, and see the comparison between these methods in mitigating real motion
Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making

6
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

The idea to guide the reconstruction using k-space uncertainty is interesting. The method section and experiments are very well. computational cost needs to be discussed.
Reviewer confidence

Confident but not absolutely certain
[Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

N/A
[Post rebuttal] Please justify your decision

N/A

Review #3

Please describe the contribution of the paper

This paper proposes a dual-domain motion correction network to correct the motion artifacts in 2D MRI slices. This method uses a k-space uncertainty module to guide the unrolled deep networks in motion corruption. The experiment results show the effectiveness of the proposed method.
Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The paper has several strengths, including:
1. The method propose a k-space uncertainty module which adjust the weight of the k-space data in data consistency module according to the degree of the corruption severity. Throughout the network, the k-space uncertainty is constantly being updated.
2. The method model the motion correction task as a dual-domain reconstruction process. The model is based on the unrolled deep network which is more interpretable.
Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
1. In Sec2, the authors state that the distribution of motion-corrupted k-space data follows a non-i.i.d. and pixel-wise Gaussian distribution. While this assumption is plausible given that many noise sources in medical imaging follow a Gaussian distribution, the authors should provide some justification for this assumption and explore other possible distributions for motion-corrupted data. It would be helpful if the authors could discuss the impact of different types of motion on this distribution. For instance, stochastic motion may fit the Gaussian assumption, but non-linear, sudden, and periodic motion in center sampling may also suit the Gaussian assumption. In addition, in other acquisition manner, single-shot or multi-shot sampling that can affect center k-space may not be suitable. The authors should consider addressing these points and possibly provide some relevant references in support of their assumption.
2. It is true that the dataset used in the experiments is relatively small, and the authors should consider testing their method on a larger dataset to validate their approach further.
3. In Sec.3, the authors should provide figures for the six freedom motion parameters in a range of light, moderate, and severe scenarios. This would help readers understand the performance of their proposed method under different motion scenarios.
4. In this manuscript, the authors propose a method for motion-corrupted MRI reconstruction using a neural network. However, as far as I know, translation motion can corrupt the k-space phase according to the equation: \begin{equation} \exp(-j\boldsymbol{k}\cdot\boldsymbol{\tau})\mathcal{F}(\boldsymbol{R}(\phi)\boldsymbol{x})\end{equation} where $\boldsymbol{\tau}$ is the motion-induced phase shift in k-space. The authors only simulate motion in the image domain, as demonstrated by the following equation: \begin{equation} \hat{y}= MF(T_\theta)x \end{equation}. This simulation approach may not be sufficient since k-space phase corruption can have a significant impact on image quality. Therefore, the authors should consider simulating translation motion in k-space and compare it with the current simulation method. By doing so, the authors can provide a more comprehensive evaluation of their proposed method’s performance under different types of motion.
5. It would be interesting to compare the proposed optimization method with an optimization-based motion estimation method. The authors should consider adding this comparison to their experiments and discuss the differences in performance between the two methods.
6. Providing the code, simulation methods, and datasets used in this study after acceptance would benefit the motion correction community. Sharing these resources allows for reproducibility of results and validates the effectiveness of the proposed method. Therefore, I recommend the authors to make their resources publicly available to the research community after acceptance.
Please rate the clarity and organization of this paper

Good
Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance

This paper’s methodology appears to be reproducible. It will be good that if the authors make the code publicly available
Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html

Please refer to the main weaknesses of the paper.
Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making

5
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

The method is interesting and the combination of the k-space uncertainty and unrolled dual-domain network is impressive. But there are also several problems existing in the current manuscript.
Reviewer confidence

Very confident
[Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

N/A
[Post rebuttal] Please justify your decision

N/A

Review #1

Please describe the contribution of the paper

The authors proposed a dual domain motion correction network, which combines k-space uncertainty (KU) and dual domain reconstruction (DDR) modules. The KU module measures the k-space uncertainty corrupted by motion, and DDR module reconstructs the motion-free MR images in both k-space and image domains under the guidance of the uncertainty estimated by KU module. The authors claim that the proposed method outperforms the existing methods in terms of PSNR, SSIM, etc.
Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

They authors propose to use k-space uncertainty to measure uncertainty of k-space corrupted by motion, which can be potentially effective and practically useful.
Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

The procedure of motion artifacts simulation is still unclear for the readers to follow. Some of the steps do not match the practical situations. (1) The authors mentioned that they simulate both in-plane and through-plane motion. The fastMRI only contains 2D multi-slice MR data, how can through-plane motion be simulated with 2D data? (2) They keep 7% of the central k-space lines, while motion may still occur when acquiring these data. (3) How are the motion artifacts simulated for fast spin echo (FSE) data?
Please rate the clarity and organization of this paper

Very Good
Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance

Nil
Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2023/en/REVIEWER-GUIDELINES.html

1.More details for motion simulation should be included for the readers to follow. 2.It would be more convincing if the authors could evaluate their proposed method(s) using data acquired with subject motion. 3.The proposed method can be evaluated with data from different types of sequences/contrasts, e.g., T1w imaging with GRE, T2w imaging with FSE, FLAIR imaging with FSE, etc.
Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making

6
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

The proposed method is novel and clearly elaborated in this paper. More details for motion simulation should be included for the readers to follow. The results from simulation study is sound, while can be further improved with acquired data with motion corruption.
Reviewer confidence

Very confident
[Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

N/A
[Post rebuttal] Please justify your decision

N/A

Primary Meta-Review

Please provide your assessment of this work, taking into account all reviews. Summarize the key strengths and weaknesses of the paper and justify your recommendation. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.

The reviewers unanimously recommended the acceptance of the paper. The meta reviewer carefully reads the reviewers’ feedback and agrees with their scores. The authors should address the mentioned issues and provide a more comprehensive evaluation of their proposed method’s performance under different types of motion, including the simulation of translation motion in k-space.

Author Feedback

N/A

back to top

Dual Domain Motion Artifacts Correction for MR Imaging Under Guidance of K-space Uncertainty