List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Jiazhen Pan, Daniel Rueckert, Thomas Küstner, Kerstin Hammernik
Abstract
Motion-compensated MR reconstruction (MCMR) is a powerful concept with considerable potential, consisting of two coupled sub-problems: Motion estimation, assuming a known image, and image reconstruction, assuming known motion. In this work, we propose a learning-based self-supervised framework for MCMR, to efficiently deal with non-rigid motion corruption in cardiac MR imaging. Contrary to conventional MCMR methods in which the motion is estimated prior to reconstruction and remains unchanged during the iterative optimization process, we introduce a dynamic motion estimation process and embed it into the unrolled optimization. We establish a cardiac motion estimation network that leverages temporal information via a group-wise registration approach, and carry out a joint optimization between the motion estimation and reconstruction. Experiments on 40 acquired 2D cardiac MR CINE datasets demonstrate that the proposed unrolled MCMR framework can reconstruct high quality MR images at high acceleration rates where other state-of-the-art methods fail. We also show that the joint optimization mechanism is mutually beneficial for both sub-tasks, i.e., motion estimation and image reconstruction, especially when the MR image is highly undersampled.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16446-0_65
SharedIt: https://rdcu.be/cVRT9
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
The authors propose an end-to-end reconstruction and registration network for MR image reconstructionn. Their proposed approach, effectively iterates reconstruction and registration steps within the same network, which allows them to train and execute them jointly.
- 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 method is interesting and the results quite impressive. The network is also extremely fast to execute. Testing is adequate for papers in this venue and all the information about training was provided.
- 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.
No major weakneses.
The authors disregard methods that could iteratively register and reconstruct, although those are computationally expensive. On this regard, the authors could test whether an iterative approach of registration-reconstruction using GRAFT and CG-SENSE (trained separately) performs similarly.
The results using elastix are worse than expected. It may be possible to improve them with more work on that end.
- 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
No data and software are provided.
The authors appear to provide sufficient details about the training of the network and reproducing it may be feasible with some effort.
- 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/2022/en/REVIEWER-GUIDELINES.html
This is very nice work. I would recommend improving the competing results with elastix for a more fair validation.
- 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
7
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This is a very interesting approach with good results
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
1
- 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 #3
- Please describe the contribution of the paper
The authors propose a learningbased self-supervised framework for MCMR, to efficiently deal with nonrigid motion corruption in cardiac MR imaging. A dynamic motion estimation was embed into the unrolled optimization, which can deliver more precise and deailed estimation.
- 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.
A dynamic motion estimation was embed into the unrolled optimization, which can deliver more precise and deailed estimation. Furthermore, the proposed method provides a motion-resolved image sequence in which all frames are motion-corrected.
- 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.
A clear description of the mathematical setting is needed.
- Please rate the clarity and organization of this paper
Excellent
- 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
good.
- 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/2022/en/REVIEWER-GUIDELINES.html
This paper is written very well. So, I recommend it for publication. However, one minor remarks: if I am nor wrong, references should be numbered consecutively in order in the text. so, please check the offical template of the MICCAI.
- 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
7
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This paper is well constructed with improved results and a good discussion.
- Number of papers in your stack
1
- What is the ranking of this paper in your review stack?
1
- 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 #2
- Please describe the contribution of the paper
The proposed work solves an important problem related to dynamic MRI data of accounting for motion compensation while reconstructing undersampled data. This method provides a framework to perform motion compensated and high quality MRI data acquired in low scan times.
- 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.
- Paper targets an important clinical problem. High spatial and temporal resolution is important for dynamic MRI scans.
- While MRI data undersampling is performed to reduce scan time, motion corruption is tackled by methods such as group registration (and of course, lower scan time implies reduced chances of motion). Conventional group wise registration methods are effective, but take a lot of time. The proposed method reduces this processing time by multiple folds.
- The most important highlight of the work is how unrolling reconstruction along with undersampled MRI reconstruction helps in improving the results both qualitatively and quantitatively.
- 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.
- Primary weaknesses of the paper have been discussed well in the text.
- Kindly mention if actual k-space data was used for training and testing or synthetic k-space data was used.
- Please rate the clarity and organization of this paper
Excellent
- 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
If possible, authors should make the code available. Replication of this work is not straightforward because of the complex implementation framework and use of in-house data.
- 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/2022/en/REVIEWER-GUIDELINES.html
- The current reconstruction method is using CG-SENSE. With the progress in DL based reconstruction, authors might want to undersampled MRI reconstruction using deep learning based techniques.
- In future work please consider testing on some prospectively undersampled data.
- Although computational metrics are important indications of the method’s performance, would recommend having radiologist ratings for future work.
- 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 work solves a very relevant clinical problem by following a well thought out framework. Authors have clearly stated the weaknesses of the method.
- Number of papers in your stack
2
- What is the ranking of this paper in your review stack?
1
- 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 #4
- Please describe the contribution of the paper
The authors effectively combine motion estimation and reconstruction in an iterative process to jointly estimate and reconstruct cardiac MR data
- 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 joint estimation of motion fields using an existing tool (GRAFT) and combining it with CG-SENSE recon.
- Evaluating the effect of acceleration compared to other existing motion modeling tools such as elastix
- 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 novelty of the method relies in assembling existing methods rather than organically developing a new idea. However, this is also useful given the challenge of non-rigid motion in cardiac MRI
- The demonstration lacks explainable AI components to trust the resulting reconstructions.
- 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
The paper meets reproducibility criteria
- 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/2022/en/REVIEWER-GUIDELINES.html
Pl. refer to the strengths and weaknesses above
- 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 novelty of the approach over all is moderate. The improvement in metrics is demonstrated over a small curated data set. The overall impact of this to prospectively accelerate data acquisition has not been shown in this study.
- Number of papers in your stack
7
- What is the ranking of this paper in your review stack?
6
- 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
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 authors propose an end-to-end reconstruction and registration network for MR image reconstruction. All reviewers reached consensus on the novelty and clarity of the work. In the final version, pleas address the few minor remarks including references and training details.
- What is the ranking of this paper in your stack? Use a number between 1 (best paper in your stack) and n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
1
Author Feedback
We would like to thank all reviewers for appreciating our work and the helpful comments. The novelty and significance of the unrolled joint motion-compensated framework are appreciated by all reviewers. We also thank the high praise in terms of the paper’s clarity and organization that reviewers gave us.
Hereafter, we would like to address the main reviewers’ comments.
Our work provides a framework that has potential for further extensions and investigations, including but not limited to:
- testing of non-iterative trained GRAFT in the proposed unrolled framework for comparison (R1),
- testing of Elastix in the proposed framework for comparison (R1),
- testing on prospectively undersampled data (R2, R4),
- using radiologist rating for results evaluation (R2) Due to the page limitation we didn’t add these experiments in the current work but we second these suggestions totally cause that’s also what we wanted to do. We will investigate this in future studies.
Regarding a more clear mathematical setting (R3): We totally agree with this suggestion. Finding a more meaningful and clear mathematical setting to solve this joint optimization is a challenging problem and this is our focus of future work.
Regarding more hyper-parameter tuning for Elastix (R1): we have already tried different settings but it is always non-trivial to find the optimal parameters for conventional methods like Elastix that generalise well for all subjects. Moreover, while the performance of Elastix estimating motion in full-sampled images is still quite okay, it has a large difficulty in motion estimation of undersampled images. We will still try it in our future work but we don’t expect a heavy performance lifting.
Regarding the novelty of the used network (R4): the major novelty of this work lies in the unrolled framework as mentioned above, while the novelty of GRAFT is its modification and application to a clinical data cohort.
Regarding the “small dataset” comments (R4): Reconstruction behaves differently from segmentation. MR CINE data acquisition is always cost- and labor-intensive to provide a reliable database with sufficient image quality. But we agree with the reviewer and acknowledge the relatively small dataset. We plan to extend this further in the future towards larger cohorts.
Regarding the reference order (R2, meta reviewer): thanks for the kind reminder. We checked again the Springer instructions section 3.9 and the used citation style follows the guideline (order by citation or alphabet are both allowed).
Regarding more training details and the code availability (R2, meta reviewer): we would add more descriptions of the training details in the paper. Regarding the code publication, we would first try to re-organize our code in a more reader-friendly way and aim to publish it.
Best regards, the authors