List of Papers Browse by Subject Areas Author List
Abstract
Individualized cortical parcellation remains challenging due to substantial inter-subject variability and the limited generalizability of group-derived templates, particularly in transitional zones where functional topography is heterogeneous. To address this, we propose STGraph-Parcel, an uncertainty-aware spatiotemporal graph learning framework that derives individualized cortical areal maps directly from vertex-wise fMRI time series. Incorporating a population-level adolescent atlas as a soft group prior, our approach preserves anatomical consistency while accommodating individual variations. Methodologically, we integrate explicit temporal modeling with spatial graph convolutions: a dual-channel Gated Recurrent Unit (GRU) integrates raw vertex signals with attention-weighted topological neighborhood representations to extract temporal embeddings. These embeddings are subsequently aggregated via Graph Convolutional Networks (GCNs) across neighbors on the cortical mesh. Evaluated on naturalistic movie-watching fMRI from the Healthy Brain Network (HBN) cohort, our method improves intra-subject functional homogeneity and test-retest reliability. Furthermore, STGraph-Parcel generates probabilistic maps and evaluates uncertainty of parcel assignment for each vertex. Using entropy-based uncertainty quantification, the model identifies high-entropy vertices as functional integration zones. Notably, these regions predominantly localize within the Somato-Cognitive Action Network (SCAN) and adjacent auditory-linguistic systems. By revealing elevated cross-network participation and developmental modulation in these SCAN zones, our study underscores their role during youth brain maturation. Ultimately, this framework advances precision functional mapping and provides a computational lens for tracking individualized neurodevelopmental trajectories in adolescent populations.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/5310_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{XueYux_UncertaintyAware_MICCAI2026,
author = { Xue, Yuxin AND Li, Wen AND Li, Jianyu AND Jin, Xin AND Dai, Jingguo AND Shen, Haozheng AND Cheng, Luqi AND Shi, Weiyang AND Zhang, Yuanchao AND Hao, Yongfu AND Zhou, Wenjie AND Zhang, Huixiong AND Zhang, Yu},
title = { { Uncertainty-Aware Spatiotemporal Graph Learning for Individualized Cortical Parcellation and Somato-Cognitive Integration in the Developing Brain } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16882},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
(1) Methodologically, the STGraph framework achieved individualized cortical parcellation and identified the most uncertain regions in parcellation as integration zones：for each vertex, they extract a dual-channel dynamics features by multi-head local self-attention block; then employ a spatial-temporal graph convolutional network with stack-GRUs to get each parcel logits and probabilistic confidence estimates; compute entropy and aggregate high-entropy vertices as group-level integration zones (2) In analysis, the integration zones identified by proposed model are highly consistent with the SCAN network and auditory-linguistic integration hubs, then analyzed its functional connectivity properties and dynamic modulation during 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.
(1) The framework combines multi-head local self-attention with spatial-temporal graph convolutional networks, and is applied to fMRI data to achieve individualized cortical parcellation and identify integration zones (2) The dual-channel signal (raw signal + attention-weighted neighborhood signal) preserves both local dynamics and context-modulated information (3) Probabilistic output + entropy computation：The entropy-based aggregation of high-uncertainty vertices interprets high-entropy vertices as “group-level functional integration zones”
- 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) Lack of comparison with other individualized parcellation methods, such as: Wang, D. , Buckner, R. , Fox, M. et al. Parcellating cortical functional networks in individuals. Nat Neurosci 18, 1853–1860 (2015). (2) Lack of ablation study: In the proposed framework, the individual contributions of multi-head attention, dual-channel GRU, graph convolution, and entropy threshold are unknown. (3) The parcellation and functional integration zones rely entirely on the natural movie-watching task; it remains unclear whether similar patterns of functional integration zones would be observed across other tasks or the resting state. (4) Codes or data are not available, which is important with respect to an atlas paper.
- 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.
(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?
This paper proposed a novel method for individualized brain atlas reconstruction. The paper is well organized and well written. The authors also provided enough evidence to show the validity of the proposed method.
- 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
Review #2
- Please describe the contribution of the paper
This paper proposes an uncertainty-aware spatiotemporal graph learning framework for individualized cortical parcellation, operating directly on vertex-wise BOLD dynamics rather than functional connectivity. Model entropy is reinterpreted as a biological signal, with high-entropy vertices delineating multimodal integration zones aligned with SCAN and auditory-linguistic hubs. The identified integration zones are supported by multiple neuroscientific validation.
- 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 offers an interesting perspective by treating model entropy as a biologically meaningful signal rather than something to suppress. 2.The integration zones are supported by multiple independent analyse and explicitly test cross-run reproducibility.
- 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 description of training is insufficient. “ Training combined cross-entropy loss, cross-session consistency objectives, and mask regularization to promote stable and biologically plausible parcellations. ” What’s the supervised label of cross-entropy loss? How to measure the cross-session consistency? 2.Group priors are mentioned but not specified. The paper states that vertex-level confidence values are “fused with a group-level probabilistic atlas prior derived from a preadolescent Brainnetome atlas,” but the fusion mechanism itself is entirely unspecified. 3.There is no quantitative comparison with existing individualized parcellation methods based on connectivity. As a result, it is difficult to assess whether vertex-wise BOLD dynamics can achieve more stable or reliable parcellations than connectivity-based approaches.
- 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?
Thorough analysis but insufficient framework details.
- 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 paper proposes STGraph-Parcel, an individualized cortical parcellation framework operating directly on vertex-wise naturalistic fMRI time series. The method combines local attention, dual-channel GRU temporal modeling, graph convolution on the cortical mesh, and a population-level adolescent atlas prior. In addition to vertex-wise parcel predictions, the model produces probabilistic outputs that are converted into entropy-based uncertainty maps, which are then used to identify putative transitional integration zones, particularly within SCAN and adjacent auditory-linguistic systems in youth.
- 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.
A major strength is the problem formulation. Individualized cortical parcellation remains important, and the use of temporally resolved naturalistic fMRI is a meaningful extension beyond static connectivity-based approaches. The framework is technically coherent, and the probabilistic output is appealing because transitional cortical territories are often poorly captured by hard atlas assignments. The biological framing around SCAN-related integration zones also adds scientific interest beyond a purely technical parcellation study.
- 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 paper does not directly compare against strong individualized parcellation baselines, which is a significant limitation for a methods paper. 2.The supervision and optimization targets are not described clearly enough, especially given the difficulty of defining ground-truth labels for individualized areal mapping. 3.The current experiments support entropy as a useful uncertainty proxy, but stronger calibration or external validation would be needed for broader biological claims about integration zones. 4.Alternative explanations such as boundary ambiguity, atlas mismatch, signal quality, or motion effects are not yet fully addressed. 5.The reported reliability evidence appears to rely mainly on within-paradigm splits rather than clearly independent retest sessions, and robustness/statistical reporting remains somewhat limited for a developmentally heterogeneous cohort.
- 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?
This paper addresses a meaningful problem in individualized cortical parcellation and proposes a technically coherent framework combining temporal modeling, graph learning, atlas guidance, and probabilistic output. The use of naturalistic fMRI and the SCAN-related analyses add scientific value. The main limitations are in validation and reporting rather than in the core idea. In particular, direct comparison with strong individualized parcellation baselines is limited, and the uncertainty-based interpretation would benefit from stronger validation. Overall, the submission is interesting and sufficiently solid to be placed slightly above the acceptance threshold. The recommendation is Weak Accept.
- 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 for their constructive feedback and for recognizing our core innovations: an uncertainty-aware spatiotemporal graph framework for individualized dynamic cortical parcellations and the validation of SCAN-related integration zones. Below we address the main concerns. R1Q1/R2Q3/R3Q1: Comparison with existing individualized parcellation methods. A: We appreciate this insightful suggestion. Unlike classical approaches relying on static functional connectivity, our key innovation lies in directly modeling dynamic vertex-wise BOLD signals to capture temporal transitions and probabilistic assignments. We agree that systematic comparison with established methods, e.g., Wang et al. (2015), will highlight these dynamic advantages. We will revise the manuscript to better position our method relative to existing individualized parcellation approaches. R1Q2: Ablation of model components. A: Our architecture was designed to address distinct spatiotemporal modeling challenges: local attention captures neighborhood-sensitive BOLD dynamics; the dual-channel design balances each vertex’s own signal with its neighborhood; the GRU models temporal dependencies; and graph convolution respects cortical topology. Together, these components enable the extraction of our entropy-based uncertainty metric. We will strengthen the explanation of how each component supports spatial context modeling, temporal dependency modeling, and topology-aware probabilistic assignment, and present full ablation analysis as future work. R1Q3/R3Q5: Dependence on movie-watching data. A: Naturalistic movie-watching provides rich multisensory stimulation, making it highly suitable for probing integrative cortical processing. We agree that cross-task and external validation will further demonstrate the framework’s broader generalizability. Since no additional task-fMRI paradigms were available in HBN, we are currently conducting experiments using HCP resting-state, task, and movie data. For the present study, we interpret the findings primarily within the naturalistic condition and will discuss its applicability and potential domain shifts in the limitation and future work section. R2Q1/R2Q2/R3Q2: Supervision labels and group prior fusion. A: We appreciate the opportunity to clarify this crucial mechanism. Our model uses the group probabilistic atlas as a soft prior rather than a hard constraint. Specifically, it is concatenated with first-layer GCN features to estimate a vertex-wise confidence map. High-confidence vertices (>0.5) are supervised by atlas-derived pseudo-labels using cross-entropy loss, while ambiguous transition zones are not forced to match hard labels. Cross-session consistency further penalizes inconsistent predictions of the same subject. Thus, the model uses reliable atlas information as weak supervision while preserving probabilistic flexibility for individual integration zones. We will detail these fusion formulations and training hyperparameters in the revised manuscript. R3Q3/R3Q4: Entropy interpretation and alternative explanations. A: We appreciate this insightful comment. A key advantage of STGraph is its ability to quantify parcellation uncertainty from temporal BOLD dynamics. We agree that high entropy requires multifaceted validation to confirm it represents biological integration rather than boundary ambiguity, motion artifacts, or noise. In this study, we evaluated cross-run reliability of candidate integration zones and provided converging evidence from multiple modalities: DTI-based structural connectivity, functional connectivity profiles, meta-analytic decoding, and participation coefficient analysis. We will revise the manuscript to tone down causal claims regarding entropy alone, emphasizing instead that the biological interpretation rests on this multi-modal converging perspective.
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.
The paper addresses an important problem in individualized cortical parcellation using naturalistic fMRI. Although reviewers noted limited baseline comparisons, incomplete ablation/training details, and the need for stronger validation of entropy-based interpretation, they agreed that the uncertainty-aware spatiotemporal graph framework is technically coherent and scientifically meaningful. The method is clearly motivated, supported by multiple validation analyses, and demonstrates biological potential.