List of Papers Browse by Subject Areas Author List
Abstract
The MRI-derived brain network serves as a pivotal instrument in elucidating both the structural and functional aspects of the brain, encompassing the ramifications of diseases and developmental processes. However, prevailing methodologies, often focusing on synchronous BOLD signals from functional MRI (fMRI), may not capture directional influences among brain regions and rarely tackle temporal functional dynamics.
In this study, we first construct the brain-effective network via the dynamic causal model. Subsequently, we introduce an interpretable graph learning framework termed Spatio-Temporal Embedding ODE (STE-ODE). This framework incorporates specifically designed directed node embedding layers, aiming at capturing the dynamic interplay between structural and effective networks via an ordinary differential equation (ODE) model, which characterizes spatial-temporal brain dynamics. Our framework is validated on several clinical phenotype prediction tasks using two independent publicly available datasets (HCP and OASIS). The experimental results clearly demonstrate the advantages of our model compared to several state-of-the-art methods.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/0919_paper.pdf
SharedIt Link: https://rdcu.be/dV1Of
SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72069-7_22
Supplementary Material: https://papers.miccai.org/miccai-2024/supp/0919_supp.pdf
Link to the Code Repository
N/A
Link to the Dataset(s)
https://sites.wustl.edu/oasisbrains/home/oasis-3/
https://www.humanconnectome.org/study/hcp-young-adult/data-releases
BibTex
@InProceedings{Tan_Interpretable_MICCAI2024,
author = { Tang, Haoteng and Liu, Guodong and Dai, Siyuan and Ye, Kai and Zhao, Kun and Wang, Wenlu and Yang, Carl and He, Lifang and Leow, Alex and Thompson, Paul and Huang, Heng and Zhan, Liang},
title = { { Interpretable Spatio-Temporal Embedding for Brain Structural-Effective Network with Ordinary Differential Equation } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15002},
month = {October},
page = {227 -- 237}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper introduces a novel framework, STE-ODE, which combines spatio-temporal embedding techniques with ordinary differential equations to model the dynamic interactions between brain structural and effective networks. This innovative approach allows for a more comprehensive understanding of how brain regions influence each other over time. One key contribution of the paper is the emphasis on interpretability. By incorporating a parameter (γ) to identify the most crucial effective connectomes, the framework provides insights into the specific brain regions and connections that play a critical role in various cognitive processes and disease states.
The study demonstrates the utility of the STE-ODE framework in clinical phenotype prediction tasks, such as gender classification, Alzheimer’s disease risk assessment, and cognitive status evaluation.
- 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.
Leveraging ordinary differential equations to model brain dynamics sets this paper apart, allowing for the illustration of dynamic causal influence strengths across fMRI scan periods, enhancing the understanding of brain function over time.
The formulation of the ODE model to capture spatial-temporal brain dynamics showcases a strong mathematical background that has been carefully derived and uncovered.
The paper provides clear experimental results that highlight the advantages of the STE-ODE model over several state-of-the-art methods, emphasizing its superior performance in capturing brain dynamics and predicting clinical phenotypes accurately. Accurately conducted validation using 5-fold cross validation as well as qualitatively described implementation details.
- 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 selection of tasks, and therefore datasets, appears to be not appropriate. The problems are either already well-solved or not clinically relevant. This raises questions about the validity and significance of the proposed method. The task of gender classification on the HCP dataset has been solved using sMRI data with superior quality. Hence, the motivation for selecting this task and evaluating the proposed method on it based on fMRI BOLD signals remains unclear. With regard to the OASIS dataset, the classification of Alzheimer’s disease has reportedly been achieved with an accuracy exceeding 95%. (Hon, M., & Khan, N. M. (2017, November). Towards Alzheimer’s disease classification through transfer learning. In 2017 IEEE International conference on bioinformatics and biomedicine (BIBM) (pp. 1166-1169). IEEE. https://arxiv.org/abs/1711.11117 ) Furthermore, from a clinical standpoint, the most compelling area of research lies in the identification of Mild Cognitive Impairment (MCI) while alzheimer’s identification is not a pressing detection problem for physicians. The relevance of the regression task for the HCP dataset is also questionable, as noted in the article the dataset comprises healthy individuals, while the MMSE, DSM scales are typically designed for the assessment of mental disorders. The appropriateness of conducting a regression analysis on these scales with a healthy population is, therefore, controversial. It remains unclear why the authors opted to undertake such an analysis with the HCP dataset.
- 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
- Please provide detailed motivation and justification for the selected tasks and datasets.
- To enhance the significance of your work, it would be valuable to evaluate the performance of your framework on the more clinically relevant task of predicting mild cognitive impairment (MCI). Given that the Oasis dataset provides an opportunity for such an assessment, it would be beneficial to leverage this resource.
- 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 primary motivation for evaluating as ‘weak accept’ is the authors’ innovative methodological approach, which is grounded in strong mathematical background and incorporates interpretable parameter to identify crucial effective connections within the brain network. However, it would be more significant to compare the performance of this framework based on more relevant tasks.
- 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
This paper presents a spatio-temporal framework with directed graph for brain network learning. Specifically, the brain-effective network is conducted, and then, the STE-ODE is designed to obtain spatial-temporal brain information. The experiments test the performance of the proposed method.
- 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 image of whole framework is clear.
- The whole layout of this paper describes well.
- The experimental results seem to be good.
- 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 definition of variants, i.e., c and b, is unclear in section 2.1.
- For fMRI data, what is node feature matrix X, and how to obtain adjacency matrix.
- For structural brain data, how to obtain optimal structural adjacency matrix.
- The adjacency matrix in fMRI data is fully-connected matrix, so how to obtain in-degree and out-degree of each adjacency matrix in Eq. (5).
- What is the motivation for Eq. (6), and what is the advantages compared to the simple combination of A^s and A^f.
- It is unclear to explain the discrete expression from Eq. (7) to Eq. (8).
- Whether the length of time segments and the number of effective networks can affect the performance of experiments or not.
- How many brain regions are selected for classification task and regression task, respectively.
- 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
Please refer to the weaknesses.
- 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?
There are few limitations can be found for the proposed method.
- 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 paper proposed a directed graph embedding layer tailored for encoding effective network under the constrains of its structural counterpart. Based on the embedding layer, the authors presented the STE-ODE framework for brain network predictions, which achieved better performance than baseline methods. They also analyzed the interpretability of the framework and identified the most important connectome changes as biomarkers.
- 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 study proposed a directed graph encoder specifically designed for capturing the causal sequences from the brain-effective networks, which captures temporal effective network representations by solving an ordinary differential equation that models the brain spatial-temporal dynamics. The framework was validated on different prediction tasks and achieved better results, providing some degree of model interpretability.
- 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.
- When performing effective connectivity analysis, model assumptions are usually made based on strong prior knowledge. In this study, effective connectivity networks have been used as features for data-driven brain network prediction task. Although the problem of information direction has been considered in the embedding layer, for example the λ was used to balance the information flow into and out of each brain node, is it sufficient to include all possible digraphs through this operation alone. And this would affect the interpretability of the model to some extent, for example, how the authors obtain different optimal λ values for different tasks, which should be further explained.
- In the part of model interpretability, the role of the direction of effective connectivity networks in the model does not seem to have been fully discussed. 3.Please check equations 7 and 8 for errors, especially those used to represent increments.
- 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?
no
- 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 the part of model interpretability, the role of the direction of effective connectivity networks in the model should be further discussed, such as advantages over undirected network (functional connections)
- 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 paper innovatively proposed a tailored directed graph embedding layer for encoding efficient networks while adhering to the constraints posed by its structural counterpart. Leveraging this embedding layer, the authors developed the STE-ODE framework for predicting brain networks, surpassing the performance of baseline methods.
- 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 gratefully acknowledge the insightful comments and questions provided by our esteemed reviewers.
Main issues. A). Selection of tasks and datasets. HCP and OASIS datasets are two publicly available resources that can be easily accessed. Since many similar studies (as referred in our experimental results) validate their methods on these two datasets, using same datasets will facilitate us to compare our method with other similar approaches. We totally agree with Reviewer 1’s suggestion to focus on MCI predictions. Currently, our work does not include enough MCI subjects from OASIS, as we are primarily processing MRI data from NC and AD subjects to construct brain networks. We will prioritize processing data from more MCI subjects and will include MCI predictions in our future studies.
B). Illustrations of the Eqs. 7 and 8. The embedding of the effective network at T+1 (i.e., G^{f}(T+1)) by using the T+1 layer is the increment of this dynamic model.
C). Model interpretability. The optimal \lambda for different tasks is determined based on parameter analysis experimental results, which are presented in our supplementary materials. The question proposed by Reviewer 4 about the discussion of information directions in the effective network is very important, which is exact the content of our next paper. Due to the page limitation, we really might not include this part in this MICCAI paper.
A few other issues. D). The definitions of c and b are the dimensions of features. We explain this shortly in our camera-ready version. E). Since we cannot obtain node feature matrix from fMRI data, we randomly initialize (using Gaussian distribution) the node feature matrix as in our previous studies. We do not include functional brain networks in this work. We utilize fMRI yielded BOLD-signal to construct effective brain networks in this work. F). The length of time segments may affect the performance. We will include the discussion on this in our future studies. G). We utilize whole brain networks (all brain regions, 82 regions for HCP data and 132 regions for OASIS data as presented in the data description section) for classification and regression tasks. Our model can automatically generate the most important brain connectomes as well as corresponding brain regions to each different prediction tasks as shown in the section 3.4. H). For the optimal structural network construction. There are different methods to construct brain structural networks, however, there is no optimal one. This conclusion was indicated in our previous study [1] which compared 9 different structural network construction methods for different prediction tasks. In this paper, HCP structural network was constructed using probabilistic tractography (FSL probtrackx) and OASIS structural network was constructed using deterministic tractography (FACT) and these two structural networks have different resolutions (82 vs. 132). We intentionally employed various methods to reconstruct brain networks to demonstrate that our new framework is independent of any particular brain structural network.
Reference 1) Title: “Comparison of nine tractography algorithms for detecting abnormal structural brain networks in Alzheimer’s disease” DOI: https://doi.org/10.3389/fnagi.2015.00048
Meta-Review
Meta-review not available, early accepted paper.