List of Papers Browse by Subject Areas Author List
Abstract
Alzheimer’s disease (AD) progression is characterized by slow, heterogeneous, and subtle changes that span decades, making transition points difficult to determine. This challenge is compounded by the complexity of longitudinal clinical data, including irregular follow-up patterns and varying observation durations that traditional survival analysis models cannot handle. We present a novel regression-based survival framework with three key innovations: (1) Robust longitudinal data handling, (2) Enhanced early-stage prediction, and (3) Flexible integration with existing models. Using a partial optimization approach for mean squared error loss, our method achieves state-of-the-art performance in AD progression prediction, particularly excelling in early-stage scenarios. Ablation studies identify the regression loss component as the key driver of improved long-term prediction capability, advancing AD prognosis and broader applications in longitudinal survival analysis.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2717_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{DaiLin_Predicting_MICCAI2025,
author = { Dai, Ling and Sun, Yiqun and Gu, Jincheng and Bao, Qinsen and Liu, Feihong and Shen, Dinggang},
title = { { Predicting Alzheimer’s Disease Progression Using a Regression-based Survival Model with Longitudinal Data } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15974},
month = {September},
page = {477 -- 487}
}
Reviews
Review #1
- Please describe the contribution of the paper
A novel regression-based loss function was proposed for survival analysis to account for irregular temporal interval and censored observations in longitudinal data analysis. Experimental results on ADNI dataset demonstrated promising performance for AD prognosis.
- Please list the major strengths of the paper: you should highlight 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.
A novel regression-based loss function to handle irregular temporal interval and censored observations.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
It would be better to provide more details regarding dataset and experimental setting. Please see comments below.
- 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.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- 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.
(4) Weak Accept — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
- Several typos need to be corrected. On page 4, it seems that the Left and Right parts of Eq.(2) were referred incorrectly, and “1/N” is missing for the mean residual error in the 1st equation. Please double check it.
- While the hyperparameters were optimized by the skopt package, it’s better to demonstrate how they affect the model’s performance.
- Details about the dataset used, e.g., number of subjects in each group (normal control, MCI, AD), ratio of right/left-censored data, mean/median time-to-progression.
- Training and testing data splits. Were the data splits stratified at the subject level?
- Integration with DeepHit enhanced the performance of the proposed regression model, but integration with Centime and DeepSurv did not, as shown in Table 1. Any insights regarding this would be helpful.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
This paper introduces a regression-based survival model designed to predict the time to Alzheimer’s disease (AD) progression. The proposed method demonstrates enhanced performance when integrated with existing survival analysis techniques, offering a complementary approach to improve predictive accuracy.
- Please list the major strengths of the paper: you should highlight 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.
- This paper presents a novel regression-based survival analysis framework.
- The proposed method achieves state-of-the-art performance, demonstrating its effectiveness on ADNI.
- It provides a flexible and extensible framework that can be integrated with existing survival analysis models, enhancing their predictive power and adaptability.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
- The introduction does not clearly highlight the advantages of the proposed method over existing survival analysis approaches, leaving the justification and motivation for the new method somewhat unclear. Specific comparisons with prior methods would strengthen the paper’s rationale.
- The methodology section requires further clarification: 1) In Section 2.1, the range of the index “i” should be explicitly defined for better understanding. 2) In Section 2.2 (top of Page 4), there is a reversal in the left and right parts of Equation 2 that needs correction. 3) It is unclear why the right part of Equation 2 equals zero—this requires additional explanation. 4) The proposed loss function is a regression loss for predicting time-to-progression (TTP), but its combination with existing regression-based survival losses in Section 2.3 lacks a clear justification. Specifically, it is unclear why adding two regression losses together is appropriate, and how this combination enhances the model’s performance or addresses potential issues in the survival analysis context.
- It would be beneficial to evaluate calibration performance in the experiments, such as using the Brier score, to complement the focus on predictive accuracy and provide a more comprehensive assessment.
- The claim that the proposed method demonstrates better predictive power at the early stage needs further clarification. Specifically, the reasoning behind the results shown in Figure 2(b) should be explained to support this assertion.
- 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 mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
- The abbreviations in Fig. 3 should appear on the same line as the caption and use the same font size, rather than being placed in a separate smaller line.
- 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.
(4) Weak Accept — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The decision for a weak accept is based on the promising nature of the proposed method, which demonstrates state-of-the-art performance and flexibility in survival analysis. However, there are several weaknesses that need to be addressed. The introduction lacks a clear comparison with existing methods, leaving the justification for the new approach unclear. The methodology section requires clarification in certain areas, such as the range of the index in Section 2.1, errors in Equation 2, and the justification for combining the proposed regression loss with existing survival losses. Additionally, while the paper focuses on predictive accuracy, the evaluation would benefit from calibration metrics like the Brier score. The claim of better predictive power at early stages also needs further explanation. Despite these weaknesses, the paper shows strong potential, and the issues can be addressed with minor revisions.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #3
- Please describe the contribution of the paper
The study proposed a novel regression-based loss for survival analysis of Alzheimer progression based on longitudinal neuroimaging data, which demonstrated improved prediction power to predict time to progression, as well as compatibility with existing survival analysis models.
- Please list the major strengths of the paper: you should highlight 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 study proposed a novel regression-based loss which handles censorship in the data through partial optimization of MSE, effectively modelling both 1) the intra-subject variation representing the additional temporal information, as well as inter-subject heterogeneity. The proposed novel loss can be used either independently, or be integrated with state-of-the-art survival analysis models as additional penalty term in an composited loss, offering commentary prediction performance improvement.
Model comparison demonstrated robustness of the proposed approach, especially over time-dependent AUC to achieve risk prediction at early stage (Figure 2b).
Ablation study demonstrated the effectiveness of each component in the model.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
Page 5: The choice of hypter-paremeter value for alpha (ratio between the bias loss and regression loss) were not stated, or hyper-parameter tuning steps were not elaborated
Page 6: Longitudinal timepoint: “The cohort was followed for 2.66 ± 2.95 years (maximum 14.07 years)” doesn’t seems right.
- 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.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- 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.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The proposed novel loss represent demonstrated a valuable addition to the to the field of deep survival analysis, and potential impact to future researches by adopting the proposed approach. The lack of shared implementation that makes the reproducibility and evaluation reduced my enthusiasm to recommend it to the top category.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
Response to Reviewer Comments:
We sincerely thank the reviewers for their constructive feedback and insightful suggestions, which have significantly improved the quality of our manuscript. Below, we provide point-by-point responses to the raised concerns:
Reviewer 1:
The paper doesn’t explain how the alpha hyperparameter was chosen or tuned. Response: We used Bayesian optimization to tune alpha and other parameters. We will clarify this in the revised manuscript.
The reported follow-up duration (2.66 ± 2.95 years, maximum 14.07) seems inconsistent. Response:
The dataset exhibits a long-tailed distribution. A few subjects had very long tracking (10–14 years), causing the large standard deviation.Reviewer 2:
Fix typos in equations on page 4 (e.g., left/right parts of Eq. 2 might be swapped; “1/N” is missing in the first equation). Response: We will fix the equations, add the missing “1/N” term, and double-check all labels.
Show how hyperparameters affect model performance.
Response: Due to manuscript length constraints, we cannot add new results. However, Hyperparameter tuning improved performance by 5–10% over default settings. Details will be shared in our code repository soon.Add dataset details.
Response: We will provide more detailed dataset discription in our final verseion.Were training/test splits done at the subject level to avoid data leakage? Response: Yes. Splits were stratified by subject ID to avoid data leakage.
Why does integrating with DeepHit improve performance but not with Centime/DeepSurv? Response: DeepHit’s ranking loss aligns well with our regression framework. In contrast, Centime and DeepSurv rely on parametric assumptions that may conflict with our regression loss. We will further discuss this compatibility issue in the revised manuscript.
Reviewer 3:
The introduction doesn’t clearly explain how your method is better than existing survival models.
Response: Our method leverages longitudinal data to enhance early-stage prediction accuracy. We explicitly models intra-subject temporal differences and integrates censoring-aware optimization, enabling estimation of latent TTP. We will refine our manuscript to clarify these advantages.- Methodology Clarifications:
- Clarify the range of index “i” (e.g., 1 to N).
Response: We will specify i ranges from 1 to N.- Fix potential reversal of left/right parts in Equation 2.
Response: We will correct the equation references.- Explain why the right part of Equation 2 equals zero.
Response: As noted by Reviewer 2, we had omitted the 1/N term in the mean residual error. With this correction, the derivation will be reworked to ensure mathematical consistency.- Justify combining regression loss with survival losses.
Response: Existing survival models estimate the survival function, whereas we directly predicts TTP. By integrating our regression loss with survival losses, we align the predicted TTP with the survival distribution’s expectation. We will expand this justification in Section 2.3.Evaluate calibration performance (e.g., Brier score).
Response: Due to conference page limits, we cannot add new experiments. However, we will instead include calibration results on our project website soon.- Clarify the reasoning behind early-stage predictive power.
Response: The improved early-stage performance stems from our model’s ability to generalize patterns from limited early observations. By modeling pairwise temporal differences across visits (via LregLreg ), the framework extrapolates disease trajectories even with sparse initial data. This aligns with the smoothness assumption in neurodegenerative progression—subjects with similar early biomarker profiles and visit patterns are inferred to have comparable TTP. We will explicitly discuss this mechanism in Section 4.2 and relate it to Figure 2(b).
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A