List of Papers Browse by Subject Areas Author List
Abstract
Autism Spectrum Disorder (ASD) is a neurodevelopmental condition marked by social and behavioral abnormalities. Functional brain networks (FBNs) are widely used in ASD diagnosis. While modular organization and hub nodes are key topological features of FBNs, existing methods fail to leverage them effectively for explainable diagnosis or to jointly model intra-module and inter-module interactions. We propose Hub-enhanced Hierarchical Graph Transformer (H2GT) to address these issues. It integrates hub-guided topological priors into a multi-scale Graph Transformer, enabling joint learning of local node interactions and global module coordination. Specifically, a Hub-Prioritized Attention Mechanism highlights hub nodes, while dedicated components model fine-grained intra-module connectivity and cross-module coordination. Experiments on ABIDE I and II show that H2GT achieves state-of-the-art diagnostic performance and identifies multi-scale biomarkers, offering deeper insights into ASD pathology.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2266_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/xiuyiqi/H2GT
Link to the Dataset(s)
ABIDE I: https://fcon_1000.projects.nitrc.org/indi/abide/
ABIDE II: https://fcon_1000.projects.nitrc.org/indi/abide/
BibTex
@InProceedings{XiuYiq_H2GT_MICCAI2026,
author = { Xiu, Yiqi AND Chen, Geng AND Huang, Yin AND Li, Xinhong AND Wen, Xuyun AND Zhao, Shijie},
title = { { H2GT: Hub-enhanced Hierarchical Graph Transformer for Explainable Diagnosis of ASD with Functional Brain Networks } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16885},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper propose a Hub-enhanced Hierarchical Graph Transformer (H2GT) to capture the fetures about the modular organization and hub nodes. It integrates hub-guided topological priors into a multi scale Graph Transformer, enabling joint learning of local node interac tions and global module coordination. And it also use the attention to explain the models, by highlighting hub nodes, and modeling fine-grained intra-module connectivity and cross-module coordination.
- 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 proposed method enables the capture of features related to modular organization and hub nodes, which may be important for classifying brain networks.
- 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 major weaknesses of the paper: 1.The absence of multi-fold cross-validation in the classification experiments raises concerns about the reliability of the reported performance metrics of the proposed methos, as a single train-test split may yield overly optimistic or pessimistic results due to data partitioning variance. 2.The model is explained in terms of hub nodes, connections within modules, and coordination between modules. However, it may not incorporate explanations based on key brain connectivity patterns (or to identify critical subgraphs) , which are commonly used in previous studies. The modular organization and hub nodes are important but may not be sufficient for graph classification models.
- 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 provide sufficient information for 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?
The proposed method may be novel, as it enables the capture of features related to modular organization and hub nodes, which may be important for classifying brain networks.
- 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 proposes H2GT (Hub-enhanced Hierarchical Graph Transformer), a novel framework for explainable diagnosis of Autism Spectrum Disorder (ASD) using functional brain networks. The main contributions include: (1) integrating topological priors into a multi-scale Graph Transformer architecture; jointly modeling fine-grained intra-module interactions and global inter-module coordination. (2) incorporates a biologically grounded hierarchical design that captures both local and global dependencies, and (3) identified hub nodes are encoded as spatial attention biases to preserve critical brain connectivity patterns, and a topology-guided mechanism simulates brain-like information flow to enhance interpretability. The work addresses important limitations in existing methods that either focus on local interactions or global dependencies but fail to effectively leverage the modular organization and hub nodes of brain networks.
- 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 proposed H2GT effectively combines topological priors with Graph Transformer architecture. The hierarchical design that separately models intra-module (via provincial hubs) and inter-module (via connector hubs) interactions is biologically motivated and technically sound. (2)Unlike many “black box” deep learning methods, H2GT provides interpretable results at three levels: (a) hub nodes identification, (b) critical ROIs within functional modules, and (c) discriminative functional modules. The findings align with established neurobiological evidence (e.g., importance of DMN, DAN, and Limbic system in ASD). (3)The method achieves state-of-the-art performance on both ABIDE I and ABIDE II datasets, outperforming 9 baseline methods including recent Transformer-based approaches
- 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)The evaluation is conducted only on ABIDE datasets (I and II). While these are standard benchmarks, additional validation on independent, multi-site datasets would strengthen the generalizability claims. (2)The paper does not compare with very recent methods that may have reported better results on ABIDE. For instance, some recent self-supervised learning approaches for brain networks are not included. (3)The paper does not provide code availability information (anonymized link or planned release). Details about hyperparameter tuning, cross-validation strategy, data split, and computational resources are limited. (4)On ABIDE II, while achieving high specificity (87.5%), the sensitivity (51.3%) is relatively low compared to some baselines. This imbalance could limit clinical utility where missing ASD cases (false negatives) has significant consequences. (5)The paper does not analyze cases where the model fails, which could provide insights into its limitations and potential improvements. (6) The paper presents a module in the figure that is never mentioned in the text. In Fig.1 (“Inter-Module Coordination Modeling”), a component named “Deep Embedding Clustering Pooling” is shown, yet no related content is found anywhere in the manuscript.
- 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
The authors should consider: (1) Providing an anonymized code repository link for reviewers. (2) Including more details on computational resources and training time. (3) Clarifying whether site information was used in the data partitioning strategy. (4) The paper presents a module in the figure that is never mentioned in the text. In Fig.1 (“Inter-Module Coordination Modeling”), a component named “Deep Embedding Clustering Pooling” is shown, yet no related content is found anywhere in the manuscript.
- 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?
The paper presents a well-motivated and technically sound contribution to the field of brain network analysis for ASD diagnosis. The H2GT framework addresses a genuine gap in existing methods by integrating topological priors with hierarchical graph Transformers. The multi-scale interpretability is a significant strength that aligns with the clinical need for explainable AI in neuroimaging. The experimental results demonstrate competitive performance, and the ablation studies validate the key components. However, the limited external validation (only ABIDE datasets), lack of code availability, and suboptimal sensitivity on ABIDE II prevent a stronger recommendation. The identified biomarkers are consistent with existing literature, which adds credibility to the method.
- 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 #3
- Please describe the contribution of the paper
This work proposes the H²GT model, which integrates prior knowledge of hub nodes in brain networks with a hierarchical graph Transformer. The model characterizes fine-grained interactions within functional network modules and global coordination across modules, aiming to address the limitation that existing methods cannot simultaneously capture local and global brain network modeling.By designing a hub-prioritized attention mechanism, the model distinguishes provincial hubs from connector hubs and assigns learnable biases, thereby enhancing feature learning in key brain regions and improving diagnostic performance. The paper identifies ASD-related biomarkers at the levels of individual nodes, hubs, and functional modules.Experiments are conducted on the ABIDE I and ABIDE II datasets.
- 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 proposed model achieves favorable performance on the ABIDE I and ABIDE II datasets, and the effectiveness of the presented method is further validated through insightful analyses and discussions.
- 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 proposed method lacks substantial novelty. Although it reports high performance metrics, the paper does not specify the number of subjects or other detailed information of the ABIDE I dataset used, which casts doubt on the reliability of the reported results. In addition, the number of comparative methods is insufficient and the methods adopted for comparison are relatively outdated, while the explanation of the model methodology is not sufficiently clear and comprehensive.
- 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 provide sufficient information for 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.
(3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The work is evaluated in terms of methodological innovation,model performance and experimental validation.
- 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
N/A
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.
I recommend an early accept for this paper. Reviewers highly praised the biologically motivated H2GT framework and its multi-scale interpretability for ASD diagnosis. The integration of topological priors into a hierarchical graph Transformer is technically sound. Although one reviewer gave a borderline score (3) citing minor concerns over baseline recency and dataset subject details, these do not undermine the core methodological novelty, solid experimental results, and strong clinical relevance of the work.
