Authors

Jiaxin Yue, Yonggang Shi

Abstract

Growing evidence from post-mortem and in vivo studies have demonstrated the substantial variability of tau pathology spreading patterns in Alzheimer’s disease(AD). Automated tools for characterizing the heterogeneity of tau pathology will enable a more accurate understanding of the disease and help the development of targeted treatment. In this paper, we propose a Reeb graph representation of tau pathology topography on cortical surfaces using tau PET imaging data. By comparing the spatial and temporal coherence of the Reeb graph representation across subjects, we can build a directed graph to represent the distribution of tau topography over a population, which naturally facilitates the discovery of spatiotemporal subtypes of tau pathology with graph-based clustering. In our experiments, we conducted extensive comparisons with state-of-the-art event-based model on synthetic and large-scale tau PET imaging data from ADNI3 and A4 studies. We demonstrated that our proposed method can more robustly achieve the subtyping of tau pathology with clear clinical significance and demonstrated superior generalization performance than event-based model.

Link to paper

DOI: https://doi.org/10.1007/978-3-031-43904-9_26

SharedIt: https://rdcu.be/dnwG4

Link to the code repository

N/A

Link to the dataset(s)

N/A

Reviews

Review #1

Please describe the contribution of the paper

The authors propose an AD subtyping according to PET data by using a graph clustering algorithm.
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.

There is a growing interest in discriminating AD subtypes according to disabilities evolution and protein depositions. To my knowledge this is the first paper doing automatically.

They propose a further smoothing after the standardized uptake value ratio.

They test on a simulated dataset to first validate the idea in a more controlled manner.
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 paper is generally good. I had hard time with the figures. Even if they are in high quality zoom is needed most of the time, a real pain in the printed version. Please consider enlarging them, or removing one, or increasing the font.
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

Dataset are from two subsets from ADNI which is publicly available, we hope to see a github with the code
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/2023/en/REVIEWER-GUIDELINES.html

The paper is generally good. I had hard time with the figures. Even if they are in high quality zoom is needed most of the time, a real pain in the printed version. Please consider enlarging them, or removing one, or increasing the font.

There is one ANDI (typo).
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

6
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

It seems an overall good written paper, some challenges with reading the images
Reviewer confidence

Somewhat 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 #5

Please describe the contribution of the paper

This paper describes an unsupervised clustering method to discover spatio-temporal subtypes of Alzheimer’s disease patients from neuroimaging data (tau-PET) without relying on a pre-defined parcellation of regions. It considers the temporal evolution of subtypes but uses only one image per patient by relying on the assumption that tau accumulation is a monotonic process. The method is validated on synthetic data and compared to another one on real data from ADNI.
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 aim of the method, which is to be independent from a ROI parcellation is interesting, especially as most studies relying on a parcellation do not investigate how dependent their method is to the particular parcellation they chose. The clustering method which evaluates if the patches are correctly sorted according to the temporal evolution is innovative and interesting. The validation on synthetic is necessary as there is no consensus on the existing Alzheimer’s disease subtypes so far. Moreover, this validation shows that the method is not sensitive to the difference of stages between different data sets: if the training data contains only “early” stages and the test data “late” stages, the performance will not decrease as the method relies on the order of patches and not their absolute intensity. The directionality between subjects that sorts patients in a cluster is also a really nice feature of the model, and could be more analyzed by the authors.
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 experiments on real data are not as convincing: there is no explanation on the hyperparameters chosen for the developed method and SuStaIn, and no experiment showing that the method is robust to changes of these hyperparameters. For example, SuStaIn 5 ROIs are chosen, and possibly the result of the method would strongly differ from the subtypes obtained by the authors by their method with another set of regions. Also, the clinical validation of the method is not convincing: to show a difference between the subtypes clinical scores specific to cognitive syndroms associated to each region should be chosen instead of the ADAS-11, ADAS-13 and MMSE. ADAS scores are the sum of a set of scores highlighting each different cognitive issues, so to spot differences between subtypes 1 and 2 the different questions should be handled separately instead of using the final sum. Differences in terms of sex, age and level of education could also be insightful. In addition, as ADNI is a longitudinal data set a better validation would be to show that the patients in a cluster progress in time as expected according to their cluster. Especially as the assumption made (tau accumulation is a monotonic process) is not supported by any reference. Finally, the obtained subtypes should be discussed according to the clinical findings to show their clinical validity. See for example DOI: 10.1186/alzrt155 or DOI: 10.1016/S1474-4422(11)70156-9
Please rate the clarity and organization of this paper

Very 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 data sets and preprocessing method are sufficiently described, but providing the code would be absolutely necessary to ensure the paper reproducibility.
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/2023/en/REVIEWER-GUIDELINES.html
A better clinical validation of the experiments on real data would be a plus but is not necessary for a conference paper. However, if this is not improved (show remarks in main weaknesses) it should be stated in the limitations of the paper that the clinical validation is limited and the described subtypes far from being validated. On synthetic data, please comment the fact that the method obtains better clustering performance on the test sets than the training sets. Finally it could help the reader to have the assumption linking tau accumulation and temporal evolution earlier in the paper to better understand how the temporal evolution is taken into account. Also if possible please add a reference to show that this assumption is clinically relevant.

Other:
- There is a word missing in the first sentence of the paper,
- In Figure 1b and 1c, please make the purple points more visible as they are really difficult to see,
- Please comment on the choice of the hyperparameters (alpha, beta and costs) and how they could affect the results. Also is there a reason to have c_1 different from the other c_i?
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?

The topic (Alzheimer’s disease subtyping) is very relevant to the community, the paper is well validated and even though the clinical assumption and validation is a bit light, this is enough for a conference paper.
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 #2

Please describe the contribution of the paper

In this work, the authors propose a Reeb graph based approach for modeling the spatiotemporal topography of tau pathology spread in AD. Unlike SuStain which relies of pre-defined anatomical regions of interest, the proposed method defines patches on the cortical surface based on the topography of tau pathology and clustering. They compare their approach to SuStain and show that both methods generate similar subtypes, but the proposed method generalized better to unseen datasets/intensity ranges.
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 well written and organized.
- The authors evaluate their approach on 2 datasets and it shows strong generalization performance.
- The clusters are identified in a more data-driven way and don’t rely on pre-defined ROIs
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 authors could provide more background on their approach to highlight the novelty of their contribution.
- It seems like one limitation of this approach could be the computation time for constructing and simplifying the Reeb graph for individual subjects’ cortical surfaces. Please discuss in the conclusion.
Please rate the clarity and organization of this paper

Very 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 authors mention choosing certain hyper-parameters such as the unpaired costs and distance matrix weights. The work would be more reproducible if they provided more detail on how these parameters were selected - are these specific to their dataset/experiments?
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/2023/en/REVIEWER-GUIDELINES.html
- Related to the first weakness, given that SuStain is the baseline method used for comparison, it would be helpful if the authors included a summary of the SuStain approach in the introduction or in a related work section. It would also be good to include a few sentences on where reeb graphs have been applied beyond citing [11].
- Please discuss any limitations of this work, what is the computation time?
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

6
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

Overall the paper is well written and the results seem promising. The paper could be strengthened by including more discussion on related work and limitations to help the reader better understand the significance of the contribution of this work.
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 #4

Please describe the contribution of the paper

This paper proposed a Reeb graph representation of tau pathology topography on cortical surfaces using tau PET imaging data.
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.

It is good to (1) use on two public dataset, (2) focus important Tau-PET is an topic, and (2) evaluate both synthetic and real data.
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) Table 1. Demographic information is too simple, more items should be included, such as Weight, MMSE, MOCA, statistical information of these subtypes should be computed seperately. (2) Comparison is too weak, only a single method published five years age was reported. (3) Lacking recent relative studies. Advance studied should be well surveyed.
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

Method is not hard, it is possible 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/2023/en/REVIEWER-GUIDELINES.html

(1) Table 1. Demographic information is too simple, more items should be included, such as Weight, MMSE, MOCA, statistical information of these subtypes should be computed seperately. (2) Comparison is too weak, only a single method published five years age was reported. (3) Lacking recent relative studies. Advance studied should be well surveyed. (4) Table 3 and Table 2 should be put into a single table.
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

3
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

Lacking recent relative studies, compared to only a single method.
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

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.

Alzheimer’s Disease subtyping is an important topic for the community. There is broad consensus among the reviewers that this is a good paper with moderate weaknesses. While the method is interesting, some minor concerns exist regarding the clinical validation of the method as well as the lack of clarity regarding hyperparameter choices.

Author Feedback

N/A

back to top

Uncovering Heterogeneity in Alzheimer's Disease from Graphical Modeling of the Tau Spatiotemporal Topography