Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Alzheimer’s Disease (AD) represents one of the most pressing challenges in the field of neurodegenerative disorders, with its progression analysis being crucial for understanding disease dynamics and developing targeted interventions. Recent advancements in deep learning and various representation learning strategies, including self-supervised learning (SSL), have shown significant promise in enhancing medical image analysis, providing innovative ways to extract meaningful patterns from complex data. Notably, the computer vision literature has demonstrated that incorporating supervisory signals into SSL can further augment model performance by guiding the learning process with additional relevant information. However, the application of such supervisory signals in the context of disease progression analysis remains largely unexplored. This gap is particularly pronounced given the inherent challenges of incorporating both event and time-to-event information into the learning paradigm. Addressing this, we propose a novel framework, Time and Event-aware SSL (TE-SSL), which integrates time-to-event and event and data as supervisory signals to refine the learning process. Our comparative analysis with existing SSL-based methods in the downstream task of survival analysis shows superior performance across standard metrics.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/3866_paper.pdf

SharedIt Link: pending

SpringerLink (DOI): pending

Supplementary Material: N/A

Link to the Code Repository

https://github.com/jacob-thrasher/TE-SSL

Link to the Dataset(s)

https://adni.loni.usc.edu/

BibTex

@InProceedings{Thr_TESSL_MICCAI2024,
        author = { Thrasher, Jacob and Devkota, Alina and Tafti, Ahmad P. and Bhattarai, Binod and Gyawali, Prashnna and the Alzheimer’s Disease Neuroimaging Initiative},
        title = { { TE-SSL: Time and Event-aware Self Supervised Learning for Alzheimer’s Disease Progression Analysis } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15012},
        month = {October},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    The authors proposed a Time and Event-aware SSL (TE-SSL), which integrates time-to-event and event information as supervisory signals to refine feature representation learning. Survival analysis on ADNI dataset shown its superior performance across C-td and IBS.

  • 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.

    Incorperating event and time-to-event information in SSL is interesting. Survival analysis in disease progression is to some extent novel.

  • 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. some definitions in this paper are vague. such as the meaning of the z_i, z_j and z_k in Fig. 1, A(i) in all equations and etc.. how to define the censored data in survival analysis? how to process the longitudinal data in SSL?
    2. in Eq. 3, does it mean to learn similar representation among scans with the same event? if so, how to learn diverse features in SSL?
    3. lack of comparison with other SSL methods.
    4. the SSL in this paper is kind of similar to y-simCLR in MICCAI 2021. More illustrations and comparisons are needed.
    5. More down streamtasks are needed to validate the effectiveness of the proposed method.
  • Please rate the clarity and organization of this paper

    Satisfactory

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission does not provide sufficient information for reproducibility.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    please see above.

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making

    Weak Reject — could be rejected, dependent on rebuttal (3)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The unsatisfactory writing and insufficient experiments make it difficult for me to accept this article.

  • Reviewer confidence

    Very confident (4)

  • [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 Time and Event-aware Self Supervised Learning (TE-SSL) method, for analyzing the progression of Alzheimer’s Disease (AD). The TE-SSL utilizes both event occurrence and time-to-event information as supervisory signals, enhancing the model’s ability to analyze disease progression. The methodology has been validated using the ADNI dataset, demonstrating superior performance over traditional SSL approaches in survival analysis tasks.

  • 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. The proposed TE-SSL integrates time-to-event and event occurrence data as supervisory signals within a self-supervised learning framework specifically tailored for Alzheimer’s disease progression.

    2. The proposed TE-SSL method is technically sound, and the experimental results support the claims made.

  • 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 idea of using supervisory signals in SSL is not new and has been explored in other domains, making the paper less impressive.
    2. It is recommended to add more figures to illustrate the method’s architecture. Also, examples of the time-to-event labels should be provided, which will make the proposed method more clear.
    3. It is not very clear how these supervisory signal work in the proposed method.
    4. It is recommended to conduct further analyses to explore why TE-SSL performs better by visualizing the learned features or detailed comparisons of cases where TE-SSL succeeds or fails.
    5. It is recommended to add further discussions about the limitations of the proposed method.
  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    See the weaknesses above.

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making

    Weak Accept — could be accepted, dependent on rebuttal (4)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
    1. Some technical details are missing.
    2. The proposed method did not compared with existing methods.
  • Reviewer confidence

    Very confident (4)

  • [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 introduced the Time and Event-aware self-supervised learning (SSL) framework, which integrates both event and time-to-event information as supervisory signals to guide the learning process of feature representations in the context of Alzheimer’s Disease progression analysis. Comparative analysis with existing SSL-based methods in the downstream task of AD survival analysis demonstrated superior performance across standard metrics.

  • 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 of adding event and time-to-event information as supervisory signals for self-supervised contrastive learning is interesting and novel.

    The paper is overall well-written and easy to follow.

  • 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. It’s not clear how the method is still self-supervised if we are adding additional supervisory signals. In that case, I wonder if TE-SSL is the right term to call the method.

    2. The values in Table 1 need to be backed by pairwise statistical comparison between TE-SSL and the ablation baselines. The slight difference might be due to randomness.

    3. The proposed method is not compared with any baselines for progression analysis other than SSL.

    4. In the paper, the authors used contrastive learning which is a particular type of self-supervised learning. This should be explicitly clarified in the paper. Else it’s misleading for the readers since there are other types of self-supervised learning models too like the Masked autoencoder.

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    See above in the weakness section for my detailed comments.

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making

    Weak Accept — could be accepted, dependent on rebuttal (4)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The idea is overall interesting and there is methodological novelty. However, statistical comparison needs to be performed to validate the better results of TE-SSL.

  • Reviewer confidence

    Somewhat confident (2)

  • [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



Author Feedback

We thank the reviewers for the constructive feedback. We present a method to learn better feature representations for survival analysis by including time and event labels during contrastive pretraining and will release code/metadata upon acceptance. We appreciate that the reviewers found our approach novel (R3/R4), interesting (R1/R3/R4), and that our experiments demonstrated superior results (R1/R3/R4) compared to no-pretraining (DeepHit) and self-supervised baselines. Below we respond to critiques raised by the reviewers

SUPERVISORY SIGNALS (R1/R4) Novelty (R1) While the use of supervisory signals in SSL has been explored in other domains (mostly in non-medical imaging), existing methods do not consider the time-dependent nature of progressive data such as Alzheimer’s Dementia. To the best of our knowledge, incorporating time labels to capture time-dependent features has not been explored within the context of survival analysis. We will further highlight this to improve the clarity of our method’s explanation

Similarity to prior work (R4) We couldn’t find a reference to y-simCLR in MICCAI21, but we believe “y-Aware InfoNCE” (Contrastive Learning with Continuous Proxy.. [MICCAI21]) was intended. It uses proxy data such as age to improve representations for classification but does not consider disease progression. While it may be somewhat similar to the E-SSL method, we believe that incorporating time information makes our proposed TE-SSL novel and unexplored

EVALUATION (R1/R3/R4) (R1/R4) Using temporal information to guide contrastive learning for progression analysis is novel, so we have no existing methods to compare against. However, we evaluated our proposed TE-SSL against DeepHit (standard approach in deep learning-based progression analysis), SSL (no supervisory signals), and E-SSL (adding supervisory signals but not used in prior progression frameworks). While other approaches use supervisory signals to guide SSL, these methods do not explicitly capture time progression and may not be suitable for survival analysis. R4 also added that more downstream tasks are required for validation, which we attribute to the limited datasets and tasks related to progression analysis

(R1) We find the suggestion regarding detailed case-by-case comparisons interesting and will include such analysis in future work. For feature-space visualization, we have already provided a t-SNE visualization of the learned features (Fig. 2)

Statistical Significance (R3) As shown in the main results (Table 1), across 3 random runs, our method improves over the existing baseline (i.e., DeepHit) by 7% for C-td and 11% for IBS. Even against SSL-based baselines, we achieve considerable improvement across both metrics. While we acknowledge that our analysis would be stronger with significance tests (which we can’t show due to MICCAI’s no experiment rebuttal policy), we believe the improvement is due to our model learning time and event features, not randomness

OTHER COMMENTS More figures (R1) Since we are relying on standard DeepHit architecture, we only presented the overall framework in Fig 1

Naming of TE-SSL (R3) We will emphasize that our framework is based on a contrastive learning objective in our final submission.

(R4) We recognize that some fundamental terms (e.g data censoring) were not explained clearly. Further, although we described some notations (e.g z_i) in our equations, they were not made clear in Fig 1. We will fix these in our final submission, also clarifying how longitudinal data is handled in SSL (each visit is a unique data point)

(R4) With Eq 3, we learn representations among scans with the same event while also strengthening alignment between scans at similar stages in development. This is enforced softly (unlike supervised learning) in SSL, allowing the framework to cluster patients with similar disease progression together

Limitations (R1) We will include discussion on limitations in our final submission



Meta-Review

Meta-review #1

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    This paper presents a Time and Event-aware Self-Supervised Learning (TE-SSL) method for Alzheimer’s disease progression analysis. The reviewers highlighted the novelty of introducing time-to-event information to contrastive learning and the good organization of the paper as strengths. As weaknesses, they noted the similar idea of using supervisory signals for SSL in other domains and the need for a more detailed description of the methodological parts. The decision was split: two Weak Accept (WA) and one Weak Reject (WR). Since the rebuttal addressed many issues, the meta-reviewer recommends accepting this paper if there is some space.

  • 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).

    This paper presents a Time and Event-aware Self-Supervised Learning (TE-SSL) method for Alzheimer’s disease progression analysis. The reviewers highlighted the novelty of introducing time-to-event information to contrastive learning and the good organization of the paper as strengths. As weaknesses, they noted the similar idea of using supervisory signals for SSL in other domains and the need for a more detailed description of the methodological parts. The decision was split: two Weak Accept (WA) and one Weak Reject (WR). Since the rebuttal addressed many issues, the meta-reviewer recommends accepting this paper if there is some space.



Meta-review #2

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    This is an borderline paper that proposes a novel method for Alzheimer’s disease progression analysis. The reviewers raised some issues that have been adequately addressed in the authors rebuttal. Therefore, I recommend accepting this paper. I encourage the authors to fully implement those comments in the final paper.

  • 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).

    This is an borderline paper that proposes a novel method for Alzheimer’s disease progression analysis. The reviewers raised some issues that have been adequately addressed in the authors rebuttal. Therefore, I recommend accepting this paper. I encourage the authors to fully implement those comments in the final paper.



back to top