List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Xiaowei Yu, Dan Hu, Lu Zhang, Ying Huang, Zhengwang Wu, Tianming Liu, Li Wang, Weili Lin, Dajiang Zhu, Gang Li
Abstract
Longitudinal infant brain functional connectivity (FC) constructed
from resting-state functional MRI (rs-fMRI) has increasingly become a pivotal
tool in studying the dynamics of early brain development. However, due to various reasons including high acquisition cost, strong motion artifact, and subject
dropout, there has been an extreme shortage of usable longitudinal infant rs-fMRI
scans to construct longitudinal FCs, which hinders comprehensive understanding
and modeling of brain functional development at early ages. To address this issue, in this paper, we propose a novel conditional intensive triplet network
(CITN) for longitudinal prediction of the dynamic development of infant FC,
which can traverse FCs within a long duration and predict the target FC at any
specific age during infancy. Targeting at accurately modeling of the progression
pattern of FC, while maintaining the individual functional uniqueness, our model
effectively disentangles the intrinsically mixed age-related and identity-related
information from the source FC and predicts the target FC by fusing well-disentangled identity-related information with the specific age-related information.
Specifically, we introduce an intensive triplet auto-encoder for effective disentanglement of age-related and identity-related information and an identity conditional module to mix identity-related information with designated age-related information. We train the proposed model in a self-supervised way and design
downstream tasks to help robustly disentangle age-related and identity-related
features. Experiments on 464 longitudinal infant fMRI scans show the superior
performance of the proposed method in longitudinal FC prediction in comparison
with state-of-the-art approaches.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16452-1_25
SharedIt: https://rdcu.be/cVRY7
Link to the code repository
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
The authors present a new Deep Learning architecture for longitudinal Functional Connectivity prediction in infants. Besides describing in detail new ideas for architecture design, they show the superiority of theirs with respect to other state of the art models.
- 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 is, as in any other Deep Learning work, the better performance when compared with other available methods. The description of the architecture is rather clear and greatly appreciate from the reader’s point of view. Personally, I think that given appropriate testing, this model shows great potential to be accepted in the clinical atmosphere given its simplicity with respect to the current competitors. Congratulations on designing from scratch an architecture that performs good.
- 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 few concerns arise when inspecting the present paper.
MAJOR COMMENTS: The so-called identity conditional module is critical for the good performance of the model, yet the description in section 2.3 is quite brief. The main idea behind its design is visible, but the inner details are not easily accessible. Furthermore, figure 2 does not provide significant help when reading the section. Figure 4 is not really a must and perhaps should be displayed in the supplemental sections. I don’t think that the result reported there should be taken into consideration when assessing the performance and validity of the model. Visual proof that the model correctly disentangles and differentiates attributes from the data is somewhat assumed whenever a model performs well. When assessing the accuracy of the predictions, only two measures are reported. Although enough to see the good performance, perhaps other network measures would proof useful for transparency and acceptance across the clinical community.
MINOR COMMENTS: As always, careful review of misspelling is advised, despite only finding small errors in the last lines of pages 2 and 7. Perhaps the reviewer misses a more extensive literature review. Even if there are not many previous works on this area, maybe larger comments on the ones already made would provide useful for anyone interested in the topic. Lastly, both pages 3 and 7 have two “alone” lines on top of the figures which might be a little bit misguiding and inconvenient for the reader. Even if space is a scarce resource, moving them to different parts of the manuscript would improve accessibility and clarity.
- 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
All the steps of the method are detailed throughout the paper easing the task of external implementations from other teams. The data comes from the BCP so easy access for reproducibility is assured. Of course, freely available and easily readable code is always appreciated by the community. As in any Deep Learning model, this last step is crucial for testing and usage from external people, so the reviewer highly encourages to publish them.
- 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 this reviewer’s opinion, the manuscript would greatly benefit from the deletion of figure 4. The space available from this deletion could be used to further expand section 2.3 (even a modification of the accompanying figure 2). More detailed literature review on other Deep Learning attempts would also be useful for any reader interested in diving into the topic of longitudinal FC prediction, regardless of the scope (infant, adult, Alzheimer, …). Reporting of other numerical measures would vastly improve the credibility of the model. Some ideas would include network measures comparisons between real and predicted.
- 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 presented work is sufficiently rigorous, transparent and engaging to be presented in a conference. It shows potential usage given further studies of its performance.
- Number of papers in your stack
4
- 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
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #3
- Please describe the contribution of the paper
The authors proposed a conditional intensive triplet network (CITN) for longitudinal prediction of the dynamic development of infant FC. This model is predicable for the common maturation pattern of FC and also maintains individual uniqueness. This model showed better performance than MLP and MWGAN methods.
- 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 prediction of brain FC maturation at individual level is hard. The authors designed a simple but high efficient conditional intensive triplet network model to capture age and individual information separately in two extractors.
- 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 influence of brain node partition should be taken into account.
- 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 reproducibility is good. The method is present clearly. The data is public 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/2022/en/REVIEWER-GUIDELINES.html
The influence of brain node partition should be taken into account. The statistical significance of comparison between the performances of three prediction models should be tested.
- 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
8
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The authors make an important and explainable network model for the challenging task of FC maturation prediction at individual level. This topic is valuable and may raise the general interests in both clinical field and fundamental research field.
- 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 #2
- Please describe the contribution of the paper
A novel conditional intensive triplet network was proposed to longitudinal predict infant FC.
- 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.
- identity and age related information could be extracted in the network.
- the predicting result looks promising.
- 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 major problem of this paper is that method parts is not clear. More detail should be give, otherwise it’s hard to follow. With such complicated designed network, ablation study is necessary, e.g. is age inf extractor really helped?
- Please rate the clarity and organization of this paper
Poor
- 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
Some hyper parameters were missing e.g. the width of hidden layer in E and G, training parameters. Thus it might be hard to reproduce this paper.
- 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 the loss design, where is the loss about encoder and ID extractor? e.g the similarity between subjects?
- It’s hard to follow how ICM module works. According to Fig2, will there be big difference if input times are 100 and 101? or is there and difference between 101 to 200? Dose the module simplify the regression problem to classification problem?
- For the compared method MLP and MWGAN, what their hyper parameters are? Since training GAN is not easy, it’s hard to conclude your method is better.
- I’m not sure if the fmri data is preprocessed well, the development pattern of two representative individuals in Fig.3 were quite different.
- 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?
The major problem of this paper is that method parts is not clear. More detail should be give, otherwise it’s hard to follow. With such complicated designed network, ablation study is necessary.
- Number of papers in your stack
4
- 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
Review #4
- Please describe the contribution of the paper
The authors propose a network to perform longitudinal infant functional connectivity prediction. In the framework, we separate age- and individual-related representations using a triplet loss and condition the extracted individual-related representation on age with an identity conditional module in order to recover the functional connectivity. The proposed method is evaluated on the baby connectome project dataset and compared to two other baselines including multi-marginal W-GAN. The proposed model outperforms the other baselines.
- 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 problem is important and well-motivated
- 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.
- In section 2.3 Identity Conditional Module, technical details are missing. It’s hard to understand the mechanism behind this module.
- To disentangle age- and identity-related information, the authors introduced the triplet loss which adding another two copies of the autoencoder architecture. This is both memory and computational expensive. Why not just reduce the mutual information between these two representations?
- The method introduced a lot of hyper-parameters such as lambda, beta, alpha. But the authors didn’t explain how they performed the searching and the effects of the hyper-parameters on the performance of the method.
- In section 3.2 Results and Visualization, the authors described the implementation details.
- The authors didn’t specify the experimental setup. For example, how train/val/test sets are divided? Are the same individuals distributed across those three sets?
- Conditional-GAN, which is a natural solution for this problem, should definitely serve as a baseline.
- 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
There’s no code link attached in the paper.
- 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
As stated above in strengths and weaknesses.
- 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?
Overall, the authors didn’t explain one module clearly in the methodology. They should also justify why they need a triplet loss instead of mutual information loss. The experimental setup is also not clear. Conditional-GAN is missing in the baseline.
- Number of papers in your stack
3
- What is the ranking of this paper in your review stack?
3
- 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 propose a network to perform longitudinal functional connectivity prediction at any specific age during infancy. Age- and identity-related information is separated during the processing and used for the prediction. The proposed framework is built on an intensive triplet auto-encoder. The proposed model outperforms state-of-the-art baselines (MWGAN, MLP) tested on almost 500 data sets from BCP.
The authors attack an important and well-motivated problem with a novel deep-learning-based framework. The major problem is that one section of the method description in particular, 2.3, is not clearly presented. More details should be given, otherwise, it’s hard to appreciate performance and reproduce the implementation of the framework. An ablation study would also be necessary given the complexity of the proposed system.
Also, a more extensive literature review would be welcome.
- 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).
3
Author Feedback
N/A