List of Papers Browse by Subject Areas Author List
Abstract
Tremendous efforts have been made to investigate stereotypical patterns of tau aggregates in Alzheimer’s disease (AD), current positron emission tomography (PET) technology lacks the capability to quantify the dynamic spreading flows of tau propagation in disease progression, despite the fact that AD is characterized by the propagation of tau aggregates throughout the brain in a prion-like manner. We address this challenge by formulating the seek for latent cortical tau propagation pathways into a well-studied physics model of the optimal mass transport (OMT) problem, where the dynamic behavior of tau spreading across longitudinal tau-PET scans is constrained by the geometry of the brain cortex. In this context, we present a variational framework for dynamical system of tau propagation in the brain, where the spreading flow field is essentially a Wasserstein geodesic between two density distributions of spatial tau accumulation. Meanwhile, our variational framework provides a flexible approach to model the possible increase of tau aggregates and alleviate the issue of vanishing flows by introducing a total variation (TV) regularization on flow field. Following the spirit of physics-informed deep model, we derive the governing equation of the new TV-based unbalanced OMT model and customize an explainable generative adversarial network to (1) parameterize the population-level OMT using generator and (2) predict tau spreading flow for the unseen subject by the trained discriminator. We have evaluated the accuracy of our proposed model using the ADNI and OASIS datasets, focusing on its ability to herald future tau accumulation. Since our deep model follows the second law of thermodynamics, we further investigate the propagation mechanism of tau aggregates as AD advances. Compared to existing methodologies, our physics-informed approach delivers superior accuracy and interpretability, showcasing promising potential for uncovering novel neurobiological mechanisms.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/2276_paper.pdf
SharedIt Link: https://rdcu.be/dV1PT
SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72069-7_47
Supplementary Material: N/A
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Hua_Uncovering_MICCAI2024,
author = { Huang, Yanquan and Dan, Tingting and Kim, Won Hwa and Wu, Guorong},
title = { { Uncovering Cortical Pathways of Prion-like Pathology Spreading in Alzheimer’s Disease by Neural Optimal Mass Transport } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15002},
month = {October},
page = {498 -- 508}
}
Reviews
Review #1
- Please describe the contribution of the paper
The author proposes an approach to modeling tau pathology spreading by formulating it as a neural optimal mass transport problem. This method combines total variation regularized optimal mass transport with Generative Adversarial Networks (GANs). The results demonstrate improved accuracy across different groups of subjects compared to existing literature methods.
- 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 methodology is well described, and experimental results demonstrate better accuracy in Alzheimer’s Disease (AD) prediction compared to some existing literature methods.
-
The formulation of tau propagation as an Optimal Mass Transport (OMT) model is a good approach, as it provides an explainable approach to understanding disease progression.
-
- 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 dataset lacks crucial demographic information such as age, sex, and MMSE scores. Additionally, details regarding the distribution of subjects across the training, validation, and test sets are necessary for transparency and reproducibility.
-
The time gap between scans is an essential consideration, as it varies between different patients and can impact disease progression. The paper should discuss how this variability is accounted for in the model, as it could influence the accuracy of predictions.
-
With N=163842, it’s unclear whether the full PET image is utilized and how noisy data is addressed in preprocessing steps. Considering the potential impact of noise on model performance, further clarification on preprocessing techniques and the suitability of graph-based models for addressing noisy data would be beneficial.
-
The paper should clearly articulate the novel contributions of its methodology compared to previous work, particularly reference [7]. Identifying and discussing specific advancements or improvements over existing approaches will strengthen the paper’s contribution to the field.
-
Fair comparison requires controlling the number of trainable parameters across all methods in the experiment. Without this control, differences in model complexity could confound the comparison of performance metrics. Ensuring consistency in model architecture and parameterization will enhance the validity of the comparison.
-
- Please rate the clarity and organization of this paper
Very 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?
The experiment section is not well described. The choice of subjects, data preprocessing should be mentioned in a more detailed way.
- 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
-
Considering the time gap between scans is crucial for forecasting tau concentration, as it directly influences disease progression and the accuracy of predictions. The paper doesn’t addressing this essential factor.
-
Additional details should be provided regarding the experiment, such as the dataset used, preprocessing steps applied, model architecture, training procedure, evaluation metrics. This will enhance transparency and reproducibility.
-
The rationale behind choosing a Generative Adversarial Network (GAN) over other network architectures should be clarified. Explaining why GANs were deemed suitable for the task at hand, and how they contribute to the model’s performance, will provide valuable insights into the design choices made in the methodology.
-
- 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 experiment is not presented in a satisfied level.
- Model is established without considering important factors in the field of pathological tau propagation.
- Reviewer confidence
Confident but not absolutely certain (3)
- [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
The author presents a variational framework for dynamical system of tau propagation in the brain.
- 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.
This topic is very interesting and important for AD. The author formulating the seek for latent cortical tau propagation pathways into a well-studied physics model of the optimal mass transport (OMT) problem, where the dynamic behavior of tau spreading across longitudinal tau-PET scans is constrained by the geometry of the brain cortex.
- 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.
Lacking comparison with the Braak stage.
- Please rate the clarity and organization of this paper
Very 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
-
This Fig2 A,D is confusing. It cannot clearly reflect the training and optimization objectives of the model.
-
According to the results of Table 1, the proposed method is the best, and whether normalization is done or not will also affect the results of MAE.
-
For the present results, we would prefer to see the results of CN, MCI and AD groups changing over the years. Whether the changes in TAU were consistent with the results of braak staging.
-
The authors first replaced W2 with Wasserstein-1 distance. Please provide the result of ablation experiments.
-
the tau flows of CN, EMCI, LMCI and AD should show in the paper.
-
- 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
Accept — should be accepted, independent of rebuttal (5)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This topic is very interesting, and well-written.
- 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 aim to model/predict the spread of the tau protein in the brain. They learn a model from longitudinal data and use physics inspired learning to model the change in distribution of tau proteins in the brain over time.
- 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 paper is very well written and address a current topic with a novel and innovative approach.
It is great to see that the apprach can handle a quite high dimension of the brain imaging data (modeling done on 160k vertices)
The explainability component of the proposed model is a plus.
Good to see validation on two datasets, although the 2nd dataset focuses on changes in cortical thickness rather than the acual tau spread.
- 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.
Main weakness is the evaluation. The evaluation focuses on ‘technical’ comparisions with other Graph Neural Networks (GNNs). However, the field has approached the question of tau spread in various ways and a validatiion using one of the other methods go a long way. Perhaps comparing to works from Garbarino et al. (https://doi.org/10.1016/j.neuroimage.2021.117980), Vogel et al. (https://doi.org/10.1038/s41467-020-15701-2) or Thompson et al. (https://doi.org/10.1162/imag_a_00089), who use different brain connecitivty models to predict tau spread.
- Please rate the clarity and organization of this paper
Very 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 provide sufficient information for reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
The method is described in a detailed way, however, access to code should be provided to ensure reproducability.
- 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
In addition to provided analyses the authors could attempt to benchmark their approach to other existing methods that go beyond different versions GNNs. As mentioned above, existing approaches use different brain connectivity graphs on which the protein spread is modeled.
- 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
Accept — should be accepted, independent of rebuttal (5)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This is a well-written paper addressing a challenging problem with a novel method.
- Reviewer confidence
Confident but not absolutely certain (3)
- [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 appreciate all the constructive comments and suggestions for all reviewers, 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. We will upload the code to Github for reproducibility in the final version.
R1 Lack comparison with Braak stage …
Thanks for the very constructive comment, we’d like to add the analysis of the comparison Braak stage in our experiment, we have separated our data according to the Braak stage and will incorporate this analysis in the final version.
Provide the ablation study…
Thanks for your comment, we will add the ablation study for W2 in the final version. W2 distance only has the minimization process.
Format and visualization problem…
Thanks for your careful feedback, we will improve Fig. 2, Table 1 and add the tau flows of CN, EMCI, LMCI and AD in the final version.
R3 …reproducibility…
We have uploaded the code to Anonymous Github for replicability and will publish it in the final version.
…provide more analyses beyond different versions GNNs…
Thanks for your very constructive feedback, we’d like to apply our method to different brain connectivity graphs (such as functional connectivity and structural connectivity), and will add a discussion section to analyze it.
R4
Lack of demographic information…
We will add detailed demographic information in the final version. The data split strategy is 6:2:2 (training: validation: testing) and we will release the code for replicability in the final version.
…the time gap between scans…
This is a good question, we did not design a separate module to handle this individual difference, because the ADNI dataset the time gap between scans in ADNI dataset is almost 1 year between two-time points. This is the limitation of our model, we will add a discussion section to discuss this variability in the final version. Thank you so much.
… provide data preprocessing steps…
Yes, we use the full PET image, we use the standard MNI Space to register all the PET images, and the number of vertices corresponds to 163842. We use the Gaussian function for denoising and will make it clear in the final version.
…the novel contributions of its methodology compared to previous work, particularly reference [7]…
Compared with Ref. [7], our work has the following novel contributions: (1) our work is based on vertices rather than regions, (2) we formulate the seek for latent cortical tau propagation pathways into a well-studied physics model of the optimal mass transport (OMT) problem, where the dynamic behavior of tau spreading across longitudinal tau-PET scans is constrained by the geometry of the brain cortex rather than inherent brain network topology (nerve fibers). (3) we use Wasserstein geodesic between two density distributions of spatial tau accumulation rather than conventional $L_1$ distance.
…the rationale behind choosing GAN…
In our work, the important reason for using GAN network architecture to implement the min-max optimization schema is that we sought to design an end-to-end deep model that allows us to include the learning component for capturing the reaction process (mapping from the observed tau concentration to the latent state) in brain cortex. We will make it clear in the final version.
Meta-Review
Meta-review not available, early accepted paper.