Authors

Angelica I. Aviles-Rivero, Christina Runkel, Nicolas Papadakis, Zoe Kourtzi, Carola-Bibiane Schönlieb

Abstract

The automatic early diagnosis of prodromal stages of Alzhei-mer’s disease is of great relevance for patient treatment to improve quality of life. We address this problem as a multi-modal classification task. Multi-modal data provides richer and complementary information. However, existing techniques only consider lower order relations between the data and single/multi-modal imaging data. In this work, we introduce a novel semi-supervised hypergraph learning framework for Alzheimer’s disease diagnosis. Our framework allows for higher-order relations among multi-modal imaging and non-imaging data whilst requiring a tiny labelled set. Firstly, we introduce a dual embedding strategy for constructing a robust hypergraph that preserve the data semantics. We achieve this by enforcing perturbation invariance at the image and graph levels using a contrastive based mechanism. Secondly, we present a dynamically adjusted hypergraph diffusion model, via a semi-explicit flow, to improve the predictive uncertainty. We demonstrate, through our experiments, that our framework is able to outperform current techniques for Alzheimer’s disease diagnosis.

Link to paper

DOI: https://link.springer.com/chapter/10.1007/978-3-031-16437-8_69

SharedIt: https://rdcu.be/cVRuU

Link to the code repository

N/A

Link to the dataset(s)

N/A

Reviews

Review #1

Please describe the contribution of the paper

In this paper, the authors propose a novel semi-supervised hypergraph learning framework for Alzheimer’s disease diagnosis. Besides,the framework allows for higher-order relations among multi-modal imaging and non-imaging data whilst requiring a tiny labelled set. The experimental results show that the proposed approach is able to outperform current techniques for Alzheimer’s disease diagnosis.
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.This paper attempted to introduce a dual embedding strategy for constructing a semantics robust hypergraph. 2.This paper presents a dynamically adjusted hypergraph diffusion model, via a semi-explicit flow, to improve the predictive uncertainty 3.This paper performed two ablations studies regarding their design and the modalities used to support the design of their technique. 4.The experimental results of‘EMCI vs LMCI’shows outperform current techniques for Alzheimer’s disease diagnosis.
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 result lacks annotation. The discussion of the differences between the work and HF [23] HGSCCA [24] HGNN [10] DHGNN [15] is insufficient and unclear. 2.The result of HF in this work is ambiguous, as different parts in the HF [23] describe it differently. 3.In section 2,the terminology and wording chosen are in parts particularly intricate, which further reduces the readability that is already affected by somewhat deviant grammar. Considering the presented approach is mathematically elaborate , the work’s reproducibility would be greatly enhanced if if it was written such that the international community could follow more easily, and the reception of a paper could likely be improved if it was easier to read. Minor revisions: 1.Problem with abbreviations ‘late mild cognitive impairment (EMCI)’，Please change it to‘(LMCI)’. 2.In ‘Fig. 3: Performance comparison for the four classes case’,the vertical axis‘HGGN’does not match what is mentioned in the text.
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

Maybe the model and results can be implemented.
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

Refer to the detailed weakness.
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?

Due to the novelty of the proposed method, the effectiveness of the results, I prefer to accept for the manuscript.
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

This paper proposed a novel self-supervised dual multi-modal embedding strategy aiming at Alzheimer’s disease diagnosis. It utilized the imaging data and the space of the hypergraph structure. Moreover, the paper introduces a diffusion model to hypergraph learning. The experimental results show the method’s good performance.
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.

Firstly, the method of this paper is innovative. Specifically, this paper uses multi-modal data (i.e. imaging and non-imaging data) to deal with the classification of Alzheimer’s disease diagnosis. And based on those data, the paper introduces a semi-supervised hypergraph learning framework. Secondly, in order to adjust the diffusion of hypergraph, the authors proposed an uncertainty hypergraph minimization of Eq. 3. Thirdly, the author’s experiments are very informative. The authors performed experiments not only for binary classification but also for multi-classification to verify the effectiveness of the proposed method.
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.

Firstly, the methodology section of the paper is not presented in enough detail to be easily understood. For example, in 2.1, the authors propose to introduce T transformations without giving detailed reasons why they should do so and what the disadvantages would be if they did not.

And for non-imaging data, the authors hold the view that creating a subject-phenotypic relation can mitigate the neglect of perturbing directly the data. This needs proof to justify. In Fig. 2, the authors do not go into detail about what each of the different colored nodes in the diagram represents and what the connecting lines between them indicate.

Secondly, the authors did not cite the most recent references for hypergraphs, e.g., literature [30].

Thirdly, the experimental part of the paper is unconvincing. The authors do not list the SEN and PPV metrics for the binary classification task in the Supplementary. Moreover, the authors’ experimental results surprisingly show that their proposed method is far ahead of other algorithms for both binary classification and multiclassification tasks (e.g., 81.69% accuracy on the AD vs NC vs EMCI vs LMCI classification task), and the authors’ explanation of this result is unconvincing.

What’s more, the author’s writing is hardly satisfactory, with frequent grammatical errors. For example, Page 1: “has show have shown” should be “has shown”. Page 2: “we introduce a a more” should be “we introduce a more”. “pLaplacian setting” should be “Laplacian setting”. Page 6: “EMCI” should be “LMCI”. “5e-2” should be “ ”. Page 7: Table1 “SEN PPV ACC” should be “ACC SEN PPV”.
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

If the code is provided, the algorithm will be easier to reproduce.
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 could make changes to address the issues shown in the weakness.
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?

Interesting paper where merits slightly weigh over weakness.
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 #3

Please describe the contribution of the paper

This paper proposes a novel semi-supervised hypergraph framework for Alzheimer’s disease diagnosis. It introduces a dual embedding strategy for constructing a robust hypergraph and a better diffusion model for hypergraph learning. Experimental results demonstrate that it outperforms other hypergraph techniques.
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 overall framework is clear and the proposed model has shown its effectiveness.
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.For the introduction, improving predictive uncertainty is not clear. 2.The proposed method is reasonable, yet hypergraph diffusion module is not clear, more in-depth analysis will be better. 3.Some typographical and grammatical improvements should be made, such as the performance of “ours” in Table 2.
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 results in this paper are easily 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

The authors need to solve the following issues: 1.For the introduction, improving predictive uncertainty is not clear. 2.The proposed method is reasonable, yet hypergraph diffusion module is not clear, more in-depth analysis will be better. 3.Some typographical and grammatical improvements should be made, such as the performance of “ours” in Table 2.
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 proposed method seems like a sound extension of jointly learning feature embeddings strategy and hypergraph diffusion module. The idea is novel. Yet hypergraph diffusion module is not clear, and some typographical and grammatical improvements should be made,
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 paper a semi-supervised hypergraph framework for diagnosis of Alzheimer’s disease. The methodology is novel and the experimental results show that the proposed approach is able to outperform current techniques.
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).

NR

Author Feedback

We thank the reviewers and meta-reviewer for their positive, insightful and very encouraging comments. All clarifications mentioned here will be included in the camera-ready.

[R1-Further discussion against SOTA techniques] We highlight that our solution has noticeable differences wrt the compared SOTA techniques. Firstly, all of them are based on the principle of [30] whilst our technique uses a different principle – a semi-explicit flow based on the Rayleigh quotient for hypergraphs. Secondly, our dual contrastive embedding strategy enforces better priors on the data, invariant to perturbations, at the image and graph level. By contrast, [23,24] use a threshold and statistical based embeddings, and in [10,15] the embeddings are assumed to be given and refined via the hypergraph structure. Importantly, all techniques only focus on the data embedding itself, whereas our technique also enforce, at the graph level, better priors for the non-imaging data whilst preserving their semantics. Thirdly, the existing techniques directly compute the maximum predicted probability from a given deep net[10,15] / ML technique[23,24] (with inherent bias), whilst our technique dynamically adjusts the uncertainty in early training stages to generate highly certain predictions.

[R1&R2 -Terminology & Details] For better clarity, we have explicitly added the clarification on how T is defined– the ratio and dropping probability strategies at the graph level, and the image transformations. Also, the explicit definition of TV_H is provided. [R2] We cited [30] as it is the core principle for current DL techniques, we also compare against the most recent works on hypergraphs DL e.g. [16].

[R2-Performance Reasoning] Unlike existing techniques that directly compute the maximum predicted probability from a given deep net/ML technique, our solution is designed to work better in two aspects wrt existing techniques. Firstly, our diffusion model seeks to mitigate the inherent network calibration problem and prediction bias in semi-supervised learning. Secondly, our solution controls the predictive uncertainty of early epochs avoiding incorrect predictions. These carefully designed advantages are not given by existing techniques.

[R3-Clarification on the Predictive Uncertainty] We refer to the predictive uncertainty to the associate confirmation bias in semi-supervised learning. That is, as the prior is solely based on a tiny labelled set, one can observe that incorrect predictions on the unlabelled set tend to have high confidence. Our solution mitigates this issue by checking the uncertainty of the predictions at early epochs to enforce the probability of the predicted labels to reflect the ground truth correctness likelihood.

[R3-Further Intuition on the Diffusion Model] Our interactive diffusion process enforces a sufficiently smooth solution to strengthen the intrinsic relation between the labelled and unlabelled data. The diffusion model seeks to propagate the tiny labelled set to the unlabelled set whilst doing an uncertainty check on the predictions. Therefore, our dynamically adjusted solution ensures the prediction correctness since early epochs avoiding propagating incorrect predictions on the hypergraph.

[R1,R2,R3-Proofreading] We thank the reviewers for pointing out typos and wordings. We rectify the typos and perform a full proof read of the submission.

back to top

Multi-Modal Hypergraph Diffusion Network with Dual Prior for Alzheimer Classification