List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Artem Razumov, Oleg Y. Rogov, Dmitry V. Dylov
Abstract
We investigate MRI acceleration strategies for the benefit of downstream image analysis tasks. Specifically, we propose to optimize the k-space undersampling patterns according to how well a sought-after pathology could be segmented or localized in the reconstructed images. We study the effect of the proposed paradigm on the segmentation task using two classical labeled medical datasets, and on the task of pathology visualization within the bounding boxes, using the recently released FastMRI+ annotations. We demonstrate a noticeable improvement of the target metrics when the sampling pattern is optimized, e.g., for the segmentation problem at x16 acceleration, we report up to 12% improvement in Dice score over the other undersampling strategies.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16446-0_73
SharedIt: https://rdcu.be/cVRUh
Link to the code repository
https://github.com/cviaai/IGS/
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
A new iterative gradient sampling algorithm was employed for finding optimal undersampling patterns in different medical tasks and applications.
- 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 idea is interesting, and the paper is well written and content rich.
- 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.
Have you ever tried to use the mixture of the two or three datasets for the training? It would be interesting to see the results.
- 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
Please provide the code.
- 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
- I find the the network training details in the Supplementary material. It it better to include some important part in the paper.
- Even though the results look promising, how to validate the learned undersampling patterns working well? Will it miss some important information?
- 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?
This paper tries to get the optimal undersampling patterns in k-space that maximize a target value function of interest in the segmentation and localization problems, and then employing a new iterative gradient method. The effort is novel. The fact that the authors have run their approach on different MR data is quite helpful. Main comments: Even though the results look promising, how to validate the learned undersampling patterns working well? Will it miss some important information?
- Number of papers in your stack
4
- 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
In the manuscript, the authors investigated MRI acceleration strategies for the benefit of downstream image analysis tasks. They proposed to optimize the k-space undersampling patterns and studied the effect of the proposed paradigm on the segmentation task using two classical labeled medical datasets, and fastMRI+ annotations. They demonstrated an improvement of the target metrics when the sampling pattern is optimized.
- 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.
Overall, the manuscript is nicely written. The Methods, Experiments, and Discussion sections are nicely presented. The Introduction also nicely summarises the motivation and related research works. The proposed methodology is interesting, and as the authors have demonstrated, retains the segmentation quality in ACDC dataset even at moderate to high acceleration factors.
- 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.
I would suggest the authors to provide additional quantitative analysis for the fastMRI dataset. My additional comments are presented below (Q#8).
- 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 authors have presented their developed tool as supplementary material, making their approach and analysis reproducible.
- 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
- There are few minor grammatical issues; I would suggest the authors to thoroughly check and revise them.
- There are too many footnotes in the manuscript, which could have been easily written inside the main body text.
- In Table 2, it is not clear which null hypotheses the p-values are referring to.
- In Eq (2), Y_hat is not defined.
- After Eq (3), c_{1,2} should be c_1 and c_2.
- Page 4: Covariance is calculated between two variables; hence, it is not clear what the authors referred to as ‘covariance of the pixel intensities’.
- 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?
Overall, the manuscript is well presented, and the proposed method is well developed and analysed.
- 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 novel way to find the optimal MRI under sampling pattern based on the downstream image processing tasks such as segmentation and pathology detection. They provide the convincing validation on large public dataset such as BRATS and fastMRI.
- 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 main strength of the paper is the development of MRI under sampling for pathology localization. Unlike other MRI under sampling based SSIM or PSNR on whole image domain, the proposed method can be more effective on specific tasks such as organ segmentation or pathology bounding box detection, which are clinically meaningful. The paper is well validated on existing public data and showed the improved performance for downstream tasks.
- 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 main weakness is the lack of explanation about optimization like IGS and LOUPE. The gradient computation on these optimizations are not clear. The convergence analysis should be included to check the stability of the method.
- 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
The optimization implementation details are needed for reproducibility. The hyper parameters are given in the supplementary files.
- 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 authors propose a new paradigm for accelerating MRI which takes account into the downstream pathology localization tasks. They need to provide the implementation details about IGS and LOUPE optimization. They can compare more advanced MRI under sampling methods like pg_mri (with greedy policy search) instead of Center or FastMRI.
- 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 major factor for my score is that they aim to find the optimal MRI under sampling pattern for downstream pathology localization instead of whole image quality. The paper shows improved performance on validation on large public data. The hyperparameters are given for reproducibility.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
2
- 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 authors studied MRI acceleration methods for the benefit of downstream image analysis jobs in the work. They proposed optimising the k-space undersampling patterns and investigated the influence of the new paradigm on the segmentation problem using two traditional labelled medical datasets and fastMRI+ annotations. They proved that when the sample pattern is optimised, the target metrics improved. The authors present a unique method for determining the best MRI under-sampling pattern based on downstream image processing tasks including segmentation and disease identification.
First, the work is indeed an interesting topic. All three reviewers post positive comments to this work. The idea is interesting, and the paper is well written and content rich (R1). Overall, the manuscript is nicely written. The Methods, Experiments, and Discussion sections are nicely presented (R2). The paper is well validated on existing public data and showed the improved performance for downstream tasks (R3). However, some concerns are: additional quantitative analysis and mixture training maybe interesting. Another shortcoming is a lack of understanding concerning optimization concepts such as IGS and LOUPE. The gradient calculation on these optimizations is ambiguous. The convergence analysis should be provided to ensure the method’s stability.
Some of my reviews:
- In the abstract, maybe it is better to report the median/quantile of the Dice score improvement. The upper bound of the Dice improvement may be misleading.
- Are the ACDC and BRaTs datasets really proper to use? I think there are lots of pre-processing has been done for BRaTs data for example.
- I found it is a bit hard to understand Figure 3, maybe use different color for different acceleration factors would be better?
- The paper is very well written and it is at the top of my stack.
Two reviewers are very confident about their reviews. Based on the review comments and my review of the work, I would say the work is important with potential value to investigate, and it is clearly above the standard of acceptance in MICCAI.
- 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 thank the reviewers for attentive reading and for their constructive feedback. We also appreciate Meta-reviewer’s detailed summary and his/her own recommendations on how to improve our manuscript. Here, we will briefly answer the primary points and will address the minor issues in the camera-ready.
1) Extra analysis and experiments: As acknowledged by reviewers, the paper is already very dense. Additional quantitative analysis and mixture training are indeed interesting but will have to wait until the follow-up paper.
2) Clarity: We will elaborate the gradient calculation on the IGS method in the proposed optimization scheme; and we only have enough room to refer to LOUPE paper without explaining it.
3) Pre-processing issues (e.g., on BRaTs and ACDC). A conventional use of the inverse FFT for the images to form a reconstruction dataset from the real-space image data is used [1]. These methods became standard in the field of compressed sensing, when there is no real k-space data.
4) We affirm that the convergence and the stability are appropriate and that there was no surprising dynamics. Convergence of the LOUPE is covered in [2].
5) Reporting in the Abstract and the colors in Figure 3 will be re-considered, as suggested, although we have to disclose that we have already gone through such search for an optimal style.
[1] Zheng, Hao et al, Cascaded dilated dense network with two-step data consistency for MRI reconstruction, NeurIPS (2019). [2] Bahadir et al, Deep-learning-based Optimization of the Under-sampling Pattern in MRI, IEEE TMI (2019).