List of Papers Browse by Subject Areas Author List
Abstract
Longitudinal PET enables regional quantification of amyloid-
β (Aβ) and tau burden in Alzheimer’s disease, yet long-horizon forecasting is challenging due to sparse, irregular follow-up and strong heterogeneity in disease stage and pace of the progression. Mechanistic progression models are interpretable but often rigid, while generic neural ODE regressors handle irregular sampling at the cost of mechanistic transparency and stability. We propose a lightweight, stage-aware mechanistic neural ODE that couples a connectome-constrained diffusion–reaction backbone with structured local residual dynamics. The model includes state-dependent diffusion modulation on a fixed structural connectome, an explicit negative-feedback channel, and a logistic-shaped residual gain, routed by a simple gating module over latent disease time and global burden with subject-specific affine time alignment. On ADNI tau-PET and amyloid-PET cohorts, we improve ROI-level forecasting on held-out subjects setting and reveal stage- and biomarker-dependent patterns in effective coupling and component contributions.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3827_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/Hongyi311/pet-progression-moe
Link to the Dataset(s)
ADNI dataset: https://adni.loni.usc.edu/
MICA-MICs dataset: https://portal.conp.ca/dataset?id=projects/mica-mics
BibTex
@InProceedings{QinHon_StageAware_MICCAI2026,
author = { Qin, Hongyi AND Fan, Yuheng AND Alam, Uazman AND Branson, Kim AND Zheng, Yalin AND El-bouri, Wahbi K.},
title = { { Stage-Aware Adaptive Reaction–Diffusion Framework for Modeling Alzheimer’s Disease Progression with Tau and Amyloid-β PET Scans } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16895},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes STARD, a stage-aware and interpretable mechanistic neural ODE framework for modeling Alzheimer’s disease progression from sparse longitudinal tau- and amyloid-PET data. The main contribution lies in bridging mechanistic modeling and data-driven flexibility while preserving interpretability. Extensive experiments on ADNI tau- and Aβ-PET datasets demonstrate improved ROI-level forecasting accuracy over both mechanistic and neural ODE baselines. Moreover, the model enables interpretable decomposition of disease dynamics, revealing stage-dependent shifts in diffusion strength and component contributions across biomarkers.
- 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.
1.The paper effectively combines a connectome-constrained reaction–diffusion model with neural ODE flexibility, achieving a good balance between interpretability and predictive power. 2.The introduction of latent disease-time alignment and stage-aware gating allows the model to explicitly account for inter-subject heterogeneity in disease stage and progression speed, which is highly relevant for Alzheimer’s disease modeling. 3.The proposed dynamic, ROI-wise diffusion modulation (dmod) significantly extends classical diffusion models by allowing disease-dependent connectivity effects rather than assuming globally fixed diffusion. 4.The decomposition into mechanistic backbone and structured local residual provides clear physiological interpretability and enables meaningful post-hoc analysis.
- 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.
1.Although the paper integrates several components, most of the underlying ideas are based on well-established prior work. For example, connectome-based reaction–diffusion models have been widely studied [1], and neural ODEs for irregular time-series modeling are already quite mature [2]. Latent disease-time modeling and temporal alignment have also been explored in neurodegenerative disease progression [3], and more recently, hybrid approaches combining mechanistic models with neural networks have been proposed [4]. As a result, the current work feels more like an integration and refinement of existing techniques rather than introducing a fundamentally new modeling paradigm. This also suggests that the related work discussion could be strengthened, as some recent developments are not fully covered. 2.Some key components, such as the structured local residual term, the negative-feedback mechanism, and the gating design, are not sufficiently motivated or explained. In particular, the connection between these components and the underlying biological processes is not clearly established, and there is limited support from prior literature. This makes the claim of interpretability less convincing. 3.The method is the main contribution of the paper and is presented in a heavily formula-driven way. However, the large number of symbols and equations makes it somewhat difficult to follow. It would help to include a short preliminaries section or a table summarizing the notations, so that readers can more easily understand the formulation. 4.Further, the paper does not provide an open-source implementation or a clear reproducibility pathway, which raises concerns about the reliability and reproducibility of the experimental results.
[1] Chaggar, Pavanjit, Jacob W. Vogel, Alexa Pichet Binette, Travis B. Thompson, Olof Strandberg, Niklas Mattsson-Carlgren, Linda Karlsson et al. “Personalised regional modelling predicts tau progression in the human brain. “ PLoS Biology 23, no. 7 (2025): e3003241. [2] Kidger, Patrick, James Morrill, James Foster, and Terry Lyons. “Neural controlled differential equations for irregular time series. “ Advances in neural information processing systems 33 (2020): 6696-6707. [3] Young, Alexandra L. , Neil P. Oxtoby, Sara Garbarino, Nick C. Fox, Frederik Barkhof, Jonathan M. Schott, and Daniel C. Alexander. “Data-driven modelling of neurodegenerative disease progression: thinking outside the black box. “ Nature Reviews Neuroscience 25, no. 2 (2024): 111-130. [4] Wen, Zheyu, George Biros, and Alzheimer’s Disease Neuroimaging Initiative (ADNI. “Aligning personalized biomarker trajectories onto a common time axis: a connectome-based ODE model for Tau–Amyloid beta dynamics. “ Medical Image Analysis (2025): 103757.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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 — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Overall, I lean towards a weak accept. The paper addresses an important problem in Alzheimer’s disease progression modeling and proposes a reasonably well-designed framework that combines mechanistic modeling with neural ODEs. The stage-aware formulation and the integration of diffusion–reaction dynamics with adaptive components are meaningful and lead to consistent improvements over several baselines. In addition, the model provides some level of interpretability, and the experimental results are fairly comprehensive. However, I do have concerns regarding the level of novelty, as many components build upon existing ideas and the contribution is more incremental. Some parts of the method are not sufficiently motivated or clearly explained, particularly in terms of their biological interpretation. The writing could also be improved for better clarity, and the lack of an open-source implementation raises minor concerns about reproducibility. Despite these issues, I believe the paper has solid technical value and could be of interest to the community, especially for researchers working on disease progression modeling and interpretable continuous-time methods. With improved clarity and stronger positioning with respect to prior work, the paper would be further strengthened.
- 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
The paper introduces a new method to track and forecast AD relevant pathology
- 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.
Focus on clinically strong targets (prognostication, staging and potentially some stratification) are meaningful and exactly where I believe these approaches could contribute.
- 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.
Sample is somewhat limited perhaps, but then there is little of this kind of data openly available.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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 — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I think the paper set a clear achievable goal and contributed meanigfully, but am less an expert on this specific topic to judge novely or accurateness.
- Reviewer confidence
Somewhat confident (2)
- [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 main contribution of this work was to develop a time survival model for predicting who will develop Alzheimer’s disease (AD) and when. The authors answered whether given a collections of features that included imaging features, could they build a survival analysis model that could predict time to AD development.
- 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 work has many strengths, and they include the following. 1.The authors include many features that are characteristic of pathological aging and Alzheimer’s disease. This includes features from imaging modalities that are traditionally used to understand disease progression in research. 2.The authors are able to show exciting results that lead to imaging features dominating the prediction since imaging metrics are generally objective features. 3.The authors use open source data which is accessible to the public. This adds to the reproducibility of the work.
- 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.
There are a few weaknesses, but only in terms of presentation. 1.Some of your figures include the “red squiggly misspell” lines in the names of the models. 2.Inside the text, there are words that are separated by dashes that should not be such as “models”. In the author’s text it has “mod-els”. 3.In section 2.5, there are several issues with the equation. It’s also missing parentheses, and the it’s not written in correct form. The authors should look over this equation once more.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/en/REVIEWER-GUIDELINES.html
Great work. I believe that, for future studies, you may want to include white matter hyperintensities (WMH) detected on T2FLAIR images (or Proton Density + T1w). WMH has been shown to be a better predictor of AD, even over hippocampal size alone. It may be worth exploring this biomarker.
- 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?
I believe researchers in the AD space have been looking for a model like this to help predict AD early so that they can help figure out treatment or lifestyle modifications that stop or slow progression. I believe the researchers still have a ways to go, but this is in the correct direction.
- 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
Author Feedback
We sincerely thank the reviewers and the AC for their careful reading, constructive comments, and positive assessment of our work.
We agree that STARD builds upon several established lines of research, including reaction–diffusion modeling, continuous-time neural models, and latent disease-time alignment. Our intended novelty is not to introduce these ideas as isolated components, but to integrate them into a constrained mechanistic-neural ODE formulation tailored to sparse longitudinal tau- and Aβ-PET progression modeling. Unlike generic neural ODE models, STARD preserves an explicitly decomposable disease-dynamics structure, thereby reducing the black-box nature of continuous-time forecasting. In addition, the state-dependent diffusion modulation and structured local residual provide controlled flexibility beyond conventional reaction–diffusion models, leading to improved forecasting performance while retaining mechanistic interpretability.
We also appreciate the reviewer’s concern regarding the biological motivation of the structured local residual, negative-feedback term, and gating mechanism. We clarify that these terms are not intended to identify independent, directly observable biological mechanisms at this stage. Rather, they are structured phenomenological channels designed to separate net local suppressive or clearance-like effects from local amplification effects, and to adapt their relative contribution across latent disease stages. In future work, we aim to further refine these modules to better disentangle heterogeneous and potentially confounded processes involved in Alzheimer’s disease progression.
We agree that, within the limited page length, presenting all mathematical details clearly is challenging. To improve readability and reproducibility, we will also release the model implementation, configuration files, and necessary preprocessing/reproducibility instructions at the camera-ready stage, subject to ADNI data-use restrictions.
We thank the reviewer for pointing out presentation issues. We will carefully correct all typographical and formatting errors in the camera-ready version, including spell-check underlines in figures, unintended word hyphenations, and equation formatting problems.
Finally, we appreciate the suggestion regarding white matter hyperintensities. While the current study focuses on modeling tau- and Aβ-PET progression, we agree that WMH and even regional brain atrophy could provide complementary information for disease progression modeling. Together, the cognition test could also be incorporated. Incorporating these additional biomarkers, while carefully handling their noise and modality-specific uncertainty, is our important direction for extending this work into a more comprehensive journal-level study.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- Please justify your decision. 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.
This paper presents a well-designed and clinically meaningful framework for modeling Alzheimer’s disease progression, with reviewers highlighting strong motivation, solid experimental validation, and useful interpretability. While the approach is largely an integration of existing ideas and thus somewhat incremental in novelty, it is technically sound and demonstrates consistent improvements over relevant baselines. Some components would benefit from clearer motivation and explanation, and the manuscript could be improved in terms of clarity and reproducibility details. Overall, the work makes a valuable contribution, and with these revisions, it meets the bar for acceptance.