List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Shengjie Zhang, Xiang Chen, Bohan Ren, Haibo Yang, Ziqi Yu, Xiao-Yong Zhang, Yuan Zhou
Abstract
Deep learning models, such as convolutional neural networks and self-attention mechanisms, have been shown to be effective in computer-aided diagnosis (CAD) of Alzheimer’s disease (AD) using structural magnetic resonance imaging (sMRI). Most of them use spatial convolutional filters to learn local information from the images, ignoring frequency domain information. In this paper, we propose a 3D Global Fourier Network (GF-Net) to utilize global frequency information that also captures long-range dependency in the spatial domain. The GF-Net contains three primary components: a 3D discrete Fourier transform, an element-wise multiplication between frequency domain features and learnable global filters, and a 3D inverse Fourier transform. The GF-Net is trained by a multi-instance learning strategy to identify discriminative features. Extensive experiment on two independent datasets (ADNI and AIBL) have demonstrated that our proposed GF-Net outperforms several state-of-the-art methods in terms of accuracy and other metrics, and can also identify pathological regions of AD. The code is released at https://github.com/qbmizsj/GFNet.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16431-6_4
SharedIt: https://rdcu.be/cVD4L
Link to the code repository
https://github.com/qbmizsj/GFNet
Link to the dataset(s)
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes a deep learning model for Alzheimer’s Disease Diagnosis from MRI images which works in the Fourier domain.
- 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.
Using Fourier networks for AD diagnosis is novel. The method outperforms a large number of competing methods. The authors present an interesting ablation study.
- 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.
Details about the competing methods are missing. There is no time comparison. The relevance of the shapely analysis is not clear because it is not done for other methods, for example CNNs.
- 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
Code will be available but the authors will not provide pre-trained models or evaluation 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
in page 2 correct “explain f_ff in details” to “in detail” In Fig 1 correct layer nrom to layer norm.
- 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?
Novelty of the method and convincing results.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
2
- 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
Not Answered
- [Post rebuttal] Please justify your decision
no change
Review #2
- Please describe the contribution of the paper
This paper proposes a 3D Global Fourier Network (GF-Net) to utilize global frequency information in predicting the Alzheimer disease (AZ). Authors used exist techniques to show the importance of frequency filter by comparing to other techniques in predicting the AZ.
- 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.
-Used public datasets like ADNI -Compare with state-of the art
- 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 idea of frequency filter is already proposed and used (ex., https://ieeexplore.ieee.org/document/7504251, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6485015/).
- Please rate the clarity and organization of this paper
Satisfactory
- 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
Could be reproducible if the code will be available since the data are public.
- 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
-Why the proposed approach only applied to Alzheimer data. We suggest testing the pipeline on many public data…,etc. The paper is almost close to clinical paper. We suggest author to note the novelty in technical aspect.
- 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
4
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Technical aspect.
- 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
3
- [Post rebuttal] Please justify your decision
Authors didn’t able to justify properly the comments we addressed.
Review #4
- Please describe the contribution of the paper
This paper proposed to utlize the Fourier transformation to catch the global information to improve the AD diagnosis, so-called GF-Net. It divides each image into several patches to meet the patch embedding operation and use a sequence of global Fourier/ inverse Fourier transform with element-wise multiplication to capture global information in the frequency domain. Multi-instance learning (MIL) strategy is utlized to randomly drop patches to augment samples for training.
- 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.
Method is described clear with good experimental results.
- 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) Motivation is not clear. Why the global information in the frequency domain is useful, What if just enhence the global information, as done by multihead attention?
(2) Not report results on MCI classificaiton. - 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
It is possible to be reproduced.
- 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
(1) It would be better to verify how much the frequency domain improve the performance. (2) It would be better to report AUC values for such binary classification. (3) It would be more useful to report results on MCI classificaiton. (4) How many patch extracted in each image? The classification results decrease along patch size, it is because of the reduce of patch counts? (5) Some typos, e.g., 0.915±0.031 should be 91.5±3.1.
- 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
4
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
No results on MCI classification. Not report AUC values.
- Number of papers in your stack
6
- 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
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
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.
Strengths: novel methodological approach, extensive comparison and ablation study
Weaknesses: unclear how the work compares to some previous work using frequency filters, novelty and motivation could be highlighted better (why is global information in the frequency domain is useful), additional results for MCI would be relevant and insightful, AUC values would improve interpretation of performance
In the rebuttal, please hightlight the novelty of the approach and motivate it better (why would it help, how does the frequency domain global information help).
- 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).
9
Author Feedback
Alzheimer’s disease (AD) is the leading cause of dementia. It is a challenging task to distinguish AD patients from healthy controls based on MRI data in clinics. To address this issue, we propose a deep learning approach with a simple but novel architecture and a single loss function, which outperforms state-of-the-art deep learning methods on two independent public AD MRI datasets. The following is our point-by-point response to reviewers’ comments.
- The idea of frequency filter is already proposed and used (reviewer 2).
Frequency filtering has existed for a long time. The novelty comes from the context of using frequency filtering. Our frequency filtering is used in a Transformer-like framework with a patch embedding, a position embedding and a residual connection. One way to see the novelty of our network is to think of replacing the multi-head attention mechanism in the Transformer by frequency filtering. The benefit is that a simple architecture is sufficient to outperform the current best method by ~2%. The whole Section 2 explains this and the extensive experiments demonstrate its advantages.
Reviewer 2 mentioned two examples of using frequency filtering, one in CNN, the other in reconstruction. These examples show that frequency filtering has been used before, but can not justify that frequency filtering in a Transformer-like architecture is not novel.
- The proposed approach is only applied to Alzheimer’s data. It is suggested to be tested on many public datasets (reviewer 2).
We only focus on AD classification using structural MRIs, as the title shows. As explained in the beginning, the significance of this task lies both in the clinical application and in gaining more insights into the disease.
- The motivation of using global information in the frequency domain is not clear (reviewer 4).
The motivation is that frequency filtering captures high-level global information compared to spatial convolution. Mathematically, multiplication in the frequency domain is equivalent to convolution in the spatial domain. However, since the learnable frequency filter can be arbitrary, it can be used to mimic convolution with a large kernel, which also captures the global information.
Interestingly, we just discovered that the idea coincides with a just accepted CVPR paper that proposes to use a large spatial kernel in the CNNs. (Xiaohan Ding et al., “Scaling up your kernels to 31x31: revisiting large kernel design in CNNs”, https://arxiv.org/abs/2203.06717).
We mentioned the motivation in the third last paragraph of Section 2.1. We also observed that if the frequency filtering is replaced with spatial convolution in our architecture, as the kernel size increases, the performance improves. We would be happy to provide more experimental results in the supplementary material if the paper gets accepted.
- No results on MCI classification are reported (reviewer 4).
We have validated our method for classifying AD patients from healthy controls in two independent public datasets, along with ablation study and interpretability analyses. For classifying pMCI and sMCI, we have done the experiments but did not include the results due to page limit. The accuracy increases by 2.5% compared to the best reported result in the literature. We plan to report these additional results in the supplementary material in the final version.
- No AUC values are reported for binary classification (reviewer 4).
We have already reported the accuracy, sensitivity, specificity, and F1-score, using all the space available in Table 1. We believe they are sufficient to show the advantages of our method. We can include the ROC curves with AUC values in the supplementary material if the paper gets accepted.
Post-rebuttal Meta-Reviews
Meta-review # 1 (Primary)
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
Points of concern are well-addressed in rebuttal.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/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).
9
Meta-review #2
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
Based on the authors’ feedback and the combined comments of the reviewers, we have decided to accept this paper. The strength is the frequency filter used in a transformer-like framework is novel.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/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).
4
Meta-review #3
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The strength is the frequency filter used in a transformer-like framework is novel. As rebuttal mentioned, the experiments on MCI as well as experiment with transformer as a comparison method could be added. Acceptance is recommended.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/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).
7