List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Tingting Dan, Minjeong Kim, Won Hwa Kim, Guorong Wu
Abstract
Amyloid-beta (A$\beta$) deposition and tau neurofibrillary tangles (tau) are important hallmarks of Alzheimer’s disease (AD). Although converging evidence shows that the interaction between A$\beta$ and tau is the gateway to understanding the etiology of AD, these two AD hallmarks are often treated as independent variables in the current state-of-the-art early diagnostic model for AD, which might be partially responsible for the issue of lacking explainability. Inspired by recent progress in systems biology, we formulate the evolving biological process of A$\beta$ cascade and tau propagation into a closed-loop feedback system where the system dynamics are constrained by region-to-region white matter fiber tracts in the brain. On top of this, we conceptualize that A$\beta$-tau interaction, following the principle of optimal control, underlines the pathophysiological mechanism of AD. In this context, we propose a deep reaction-diffusion model that leverages the capital of deep learning and insights into systems biology, which allows us to (1) enhance the prediction accuracy of developing AD and (2) uncover the latent control mechanism of A$\beta$-tau interactions. We have evaluated our novel explainable deep model on the neuroimaging data in Alzheimer’s Disease Neuroimaging Initiative (ADNI), where we achieve not only a higher prediction accuracy for disease progression but also a better understanding of disease etiology than conventional (“black-box”) deep models.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43898-1_7
SharedIt: https://rdcu.be/dnwAE
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 a predictive tool based on a combination of PDEs and machine learning approaches. The idea is to combine the initial values of AB and tau proteins to predict disease evolution.
- 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 to evaluate combined Abeta and tau protein is really interesting.
- 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 combination of PDEs for explicability of neural networks is interesting but has few conceptual pitfalls. First, it is nice that you expanded the RDM framework to handle 2 proteins but why expanding this model? There are more easy to handle spread model (e.g. the network spreading of Raj et al. which is one of the two most used spreading model for neurodegeneration). Then, it is not clear why using a PDEs to achieve explainability in neural networks, when you have already all information from the PDEs on where the depositions are.
Then, the comparison is given in terms of tau prediction, it is not clear where is category classification.
Lastly, all comparisons are with RDM or GCN variation, we know that the state of art model is the epidemic spreading of Itturia-medina, you should compare with that.
- 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
Data are from the ADNI dataset, which is highly heterogenous with different subtrial (ADNI2, ADNI3, ADNIGo)…. It is not clear how you selected these. Also you selected the patients with enough PET, but not showing from which subgroups.
- 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
Concerns about the comparison to state-of-art are mentioned above.
Abstract:
The physiological implications are “very abstract”. What do we achieve by this more advanced interaction prediction of the two proteins?
What do you mean “we selected 126 cohort”? You mean 126 patients?
Did you check the distance of AB and tau PET acquisition? Some ADNI data have even more than 1 month difference between PET and structural, not so sure about tau-PET and AB-PET
Minor:
Introduction: With all enourmouse literature about connectivity and spreading on AD, why do you cite a general paper of Bassett in sentences about AD? Can’t you find anything more suitable?
Capitalization in the references
- 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?
Missing details and comparison with state-of-art
- 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 #6
- Please describe the contribution of the paper
The authors introduce a novel framework incorporating the evolving biological process of Aβ cascade and tau propagation into a closed-loop feedback system constrained by region-to-region white matter fiber tracts in the brain. They also develop an explainable deep learning model based on the newly formulated RDM, demonstrating promising results in predicting AD progression and diagnosing the disease on the ADNI dataset.
- 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.
(1) a novel, explainable machine learning initiative to investigate the mechanism of Aβ-tau interaction in Alzheimer’s disease. They have used an RDM-based deep model to study the prion-like propagation mechanism of tau aggregates and their association with clinical manifestations in AD. This tailored deep model significantly improves the prediction accuracy of developing AD and provides new insights into the pathophysiological mechanism of disease progression using a data-driven approach. (2)an optimal constraint in the vanilla RDM, which has been well-studied in neuroscience, to improve the model’s performance further.
- 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) The author asserts that”Although the graph attention technique [30, 20] allows us to quantify the contribution of each node/link in predicting outcome, its power is limited in dissecting the mechanistic role of Aβ-tau interactions, which drives the dynamic prion-like the pattern of tau propagation throughout the brain network.” A more convincing argument could be made by adding a detailed comparison and explanation of whether the weights of node connections in graph neural networks would yield results similar to those shown in Figure 4. It would also be beneficial if the author could provide visualizations from GCN or similar models and then elaborate on their comparison and the results in Figure 4.
(2) Low novelty. The theory is derived from [17], Different from [7] introduces the linear quadratic regulator[2]
- 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
If the code and specific pre-processing and experiments are open, there is hope.
- 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
Is it possible to compare the visualization between graph neural network nodes such as GCN? or Additional evidence supports the claim that the power of graph neural networks is limited in dissecting the mechanistic role of Aβ-tau interactions?
- 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 methodology presented is intriguing; however, the author’s motivation lacks sufficient supporting arguments.
- Reviewer confidence
Somewhat 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 #7
- Please describe the contribution of the paper
Authors in this paper propose a novel deep learning framework to detect developing Alzheimer’s disease leveraging amyloid-beta cascades and tau propagations instead of using them as independent variables. In addition, they also show this approach is explainable in terms of highlighting the most vulnerable brain regions to AD progression. The method builds on Neuro-RDM [7] method and the proposed method outperforms included baselines for binary classification.
- 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 described in the paper is more pathophysiologically related to studying AD progression as it is built on Neuro-RDM model. Additionally the framework also models amyloid-beta and tau interactions which are known to be of interest in understanding AD progression.
The additional explainabilty module of the system is simple and straightforward which most importantly aligns somewhat with existing clinical findings.
The paper is nicely written and easy to follow for someone which decent background in covered concepts.
- 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.
Page 6 mentions the proposed method is evaluated on a binary classification task collected from 4 categories of diagnosis status. I am not sure why authors selected to merge CN with EMCI and LMCI with AD, as it is known the progression trajectories could be different for all 4 groups based on follow-up recorded diagnosis.
Is the proposed framework somewhat limited to binary classification? I believe it should not be but if so, might be relevant to the point above.
- 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 is aligned with responses in the reproducibility response.
- 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
Overall the paper is very well written and explained and the results are good. I would only suggest to explain more, as pointed above, more on the categorization of 2 classes included in experiments.
- 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 paper is well written and explained to someone who has enough understanding of AD progression using amyloid-beta and tau accumulation. The highlighted points above are minor points that can be clarified but the proposed method can be studied further and extended to improve AD prediction accuracy.
- 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
6
- [Post rebuttal] Please justify your decision
My decision still remains as “accept” for this manuscript.
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.
There is some variability in the reviewers’ scores, which reflect concerns across a range of topics. While the reviewers agree that the idea of combining abeta and tau information is interesting, there is a number of concerns. First, the novelty of the paper seems limited. Second, and more importantly, it is not clear why the proposed approach was selected. The authors should clarify their motivation and underlying rationale. Third, and related to the previous, the authors seem to fail to provide appropriate context by citing relevant literature. Fourth, the authors should clarify choices regarding baselines and the grouping of CN with EMCI and LMCI and AD. Regarding the baselines, it seems that relevant popular modules such as the ones by Itturia-medina or Raj were not considered, while the grouping seems erroneous in itself as clinically distinct groups are merged together. Additionally, graph attention should be discussed further. Lastly, code should be available to enhance reproducibility.
Author Feedback
We appreciate all the constructive comments and suggestions. We extend our special thanks to the meta-reviewer for summarizing the comments, which greatly facilitated our efforts in addressing the critiques effectively. We are committed to incorporating all the valuable feedback into the final version of our paper. The code has been released in Anonymous Github.
Meta-reviewer:
Limited novelty: We would like to emphasize our novelties as follows. First, the technical contributions include 1) introducing the optimal control mechanism for closed-loop feedback RDM (reaction-diffusion model) system, 2) extending the analytic RDM solution for individual subjects to a learning-based framework for capturing population behaviors, and 3) integrating the interaction between system (tau) and environment (Aβ) into deep learning. Second, our work is the first to uncover AD pathology interactions using an explainable deep model, linking deep neural networks with complex systems from physics. This enables the design of explainable models capturing nonlinear dynamics of Aβ-tau interactions, offering insights into underlying mechanisms. By harnessing the combined strengths of deep learning and fundamental principles from physics, our model offers a fresh perspective for studying biological systems and holds potential for broader applications within the field of AD research.
Motivation and underlying rationale are not clear: Current ML methods for disease diagnosis often focus on fusing multi-modal data without delving into the underlying biomarker interactions. In our work, we address the intricate relationship between Aβ and tau in AD, where Aβ accumulation triggers tau tangles, and aggregated tau promotes Aβ aggregation. We propose a closed-loop RDM with an LQR constraint to model this feedback loop. Additionally, we employ GNN for learning the model parameters. Our approach not only improves prediction accuracy but also sheds light on the dynamic Aβ-tau interaction driving disease progression, opening new avenues for neuroscience research using data-driven methods.
Missing relevant work: Due to the page limit, we have to focus on RDM and machine learning part of this interdisciplinary work. We will include all related work and comparisons in final version. We compared the result of network diffusion model (NDM) by Raj 2012. The MAE of predicting tau is 0.11±0.02 by NDM, compared to 0.02±0.02 by our method. Also, we will include the comparison with GAT in the final version (e.g., 0.81±0.03 by GAT compared with 0.84±0.05 by our method in prediction AD conversion using Aβ+tau).
Clarify the choice of clinical groups: We apologize for the confusion. The precise description for the binary classification is the prediction of conversion of AD. Since the clinical symptom is not onset until converting from EMCI to LMCI, we consider CN+EMCI as ‘non-convert’ and LMCI+AD as ‘converted’ group. We will rephrase this in final version.
R1 Thank you for the constructive comments. Very helpful. We will definitely correct them (typos, references, etc.) in final version. Epidemiology models often rely on existing biological knowledge, limiting novel insights from data. This work explores disease progression mechanisms using data-driven approaches. As a proof-of-concept, we combine control theory with a feedback RDM model, optimizing parameters through machine learning. By integrating PDEs with neural networks, we introduce interpretable components that reveal underlying model mechanisms. Physical equations serve as constraints, enhancing our understanding of prediction generation. Since AD is a neurodegenerative disease with long progression, it is reasonable to assume the structure and pathology remain stable in a short time period (e.g., 6 months). In our experiment, we match the scanning time to ensure the time gap is less than 3 months.
R6 & R7 Thank you for your comments. We have addressed the concerns in response to MetaReviewer.
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.
The authors have clarified some of the main questions raised during the review process. The paper presents interesting methodology and sufficient novelty. Overall, its merits weigh over its weaknesses.
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.
After Rebuttal
- I don’t agree with the authors on the explainability part but I think the issues are addressed.
- The model choice seems to be an overkill as pointed out by the reviewers but perhaps as proof of concept it is ok.
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.
Although the authors tried to address all the concerns raised by the reviewers, I don’t think there are enough space to include “all related work and comparisons in final version”. Considering the limited novelty, I recommend reject.