List of Papers Browse by Subject Areas Author List
Abstract
EEG-based emotion recognition is vital for patients who are unable to express emotions normally through physical or verbal means.It can provide essential support for their emotional expression and rehabilitation. EEG signals are highly non-stationary, and there is significant variability in emotional expression among individuals. The Graph Convolutional Network (GCN) has shown excellent performance in EEG signal feature extraction, but their accuracy in cross-subject scenarios remains unsatisfactory. In this paper, we propose a Various Attention Mechanism Graph Convolutional Network with Multi-Source Domain Adaptation (VAG-MSDA) model for cross-subject EEG emotion recognition. VAG extracts features through the GCN with various attention mechanism to capture the emotional cognitive attributes of the graph structure in spectral, local, and global spatial domains, ensuring the richness and stability of feature information while reducing redundancy. Additionally, MSDA is used to align the feature distributions and classifiers among different individuals, further enhancing the model’s generalization ability. Experiments were conducted on the SEED and SEED-IV datasets. The results demonstrate that the proposed VAG-MSDA model achieves significant performance improvements and reaches state-of-the-art performance levels on the SEED-IV dataset. Our code is open-sourced at https://github.com/e6ut/vag-msda.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2532_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/e6ut/vag-msda
Link to the Dataset(s)
SEED dataset: https://bcmi.sjtu.edu.cn/home/seed
SEED-IV dataset: https://bcmi.sjtu.edu.cn/home/seed
BibTex
@InProceedings{ShiShu_Various_MICCAI2025,
author = { Shi, Shuo and Zheng, Xulei and Wu, Taiyi and Hao, Xiaoke},
title = { { Various Attention Mechanism Graph Convolutional Network with Multi-Source Domain Adaptation for Cross-Subject EEG Emotion Recognition } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15971},
month = {September},
page = {680 -- 690}
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors proposed a novel model named VAG-MSDA for domain adaptation in EEG emotion recognition task. In their VAG-MSDA, they introduced several advanced techniques, such as the graph-based features, multi-domain attention module and multi-feature fusion strategy, to extract multi-view (spectral and spatial) and multi-scale (local and global spatial) features from non-stationary EEG signal. The evaluation results showed that their VAG-MSDA significantly improves accuracy on both the SEED and SEED-IV 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.
- This work introduced many techniques from advanced work. For example, they introduced the multi-view feature and attention module from reference [17]; introduced the graph-based features inspired from reference [5]; introduced dual-graph attention from references [5] and [13]. Among these, the dual-graph approach is particularly interesting and may represent a novel strategy. The authors show that combining these methods can result in significant improvement.
- The authors compare with many baseline methods (12 methods) in their experiments, including both supervised learning and transfer learning. The comparison experiment is very solid.
- 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 authors used a large number of symbols to explain their methods. However, the symbol management is disorganized. For example, all the GCN, SSAFE and MGAFE have the activation function, and the authors use σ, σ and LeakyRelu to represent them, respectively, it’s very confusing; m=softmax(Λ)h is used to represent the intermediate output of MGAFE, but also represent the numbers of source samples.
- It is hard to clearly identify the main contributions of this work, particularly how they differ from reference [17]. For example, the overall structure of their VAG-MSDA appears quite similar to S2A2-MSDA; the SSAFE module also seems identical to the one presented in [17]; and the loss function closely similar that in [17] as well. However, their main improvements, such as the introduction of GCN features, the fusion of F’, F’’ and F’’’, and the use of dual-graph features, are intertwined with existing methods, making it difficult to distinguish the novelty of their approach.
- Although the title and abstract emphasize the use of a GCN structure, the authors only adopt a simple one-layer GCN for feature extraction. Moreover, the ablation study does not evaluate whether GCN plays a decisive role in the model’s performance. It raises questions about its actual contribution to the improvements.
- 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.
- 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/2025/en/REVIEWER-GUIDELINES.html
- See the weakness, which is my major concern.
- If possible, please include your insights or comments based on the experimental results, analyzing why the proposed methods lead to performance improvements, rather than simply presenting the results. Such analysis is also crucial for helping readers understand your work.
- There are many typos throughout the manuscript, and some are severe enough to affect the clarity and comprehension of the content. For example, title of Figure 1 should be “The overall framework of VAG-MSDA”, while “The overall framework of SSAFE” for Figure 2.
- 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 — could be 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 dual-graph fusion method is somewhat interesting. In addition, this work incorporates several advanced techniques into their VAG-MSDA framework and evaluates them against a comprehensive set of baselines. However, the clarity of presentation, the disorganized description of the methodology, and the lack of analysis regarding the role of GCN as well as the absence of insights or comments on the results are concerning.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
I believe this work presents a certain level of novelty and contribution. The authors have provided reasonable clarifications during the rebuttal regarding my earlier concerns on the network design, including the use of GCN and the interactions between different modules in the proposed method. My remaining concerns primarily lie in the clarity of the manuscript. These include numerous typos, inconsistent or unclear symbols or notation, and a disorganized presentation of results, which makes it difficult to clearly identify the main contributions. However, the authors have stated in the rebuttal that they will carefully revise the manuscript to address these issues. Based on this response, I am willing to trust their revision and recommend acceptance.
Review #2
- Please describe the contribution of the paper
The paper proposes a VAG-MSDA model that combines a Various Attention Mechanism Graph Convolutional Network (VAG) and Multi-Source Domain Adaptation (MSDA) for cross-subject EEG emotion recognition. The model extracts features using GCN and various attention mechanisms while aligning feature distributions and classifiers across domains to reduce individual differences. Experiments on the SEED and SEED-IV datasets demonstrate improved performance over state-of-the-art methods.
- 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 model achieves significant performance improvements on both SEED and SEED-IV datasets, with ablation studies validating the effectiveness of each module.
- The combination of VAG and MSDA provides a structured approach to addressing cross-subject variability in EEG emotion recognition.
- 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.
- Lack of novelty. The integration of attention mechanisms and multi-source domain adaptation is not novel. Prior works have explored similar concepts in domain adaptation.
- The design of the attention modules (e.g., SSAFE and MGAFE) closely resembles existing approaches like Spectral-Spatial Attention Alignment [1], without clear discussion and citation.
- The absence of a comprehensive survey of existing methods in multi-source domain adaptation and attention methods. such as [2][3][4].
[1] Spectral-Spatial Attention Alignment for Multi-Source Domain Adaptation in EEG-Based Emotion Recognition [2] Attention guided multiple source and target domain adaptation [3] Multiple adaptation network for multi-source and multi-target domain adaptation [4] Multi-source Multi-target Domain Adaptation Based on Evidence Theory
- 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.
- 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/2025/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 — could be rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The details are refered to the weaknesses.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Reject
- [Post rebuttal] Please justify your final decision from above.
The novelty is limited as most components of the proposed network structure are borrowed from existing works.
Review #3
- Please describe the contribution of the paper
1) This paper propose a Various Attention Mechanism Graph Convolutional Network with Multi-Source Domain Adaptation model (VAG-MSDA) to ensure sufficient feature extraction while maintaining low redundancy in the features and to overcome the impact of individual differences. 2) This method combines the strengths of GCN in handling graph-structured features and the superior performance of various attention mechanisms.
- 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 method focus on spectral domain information, local, and global spatial information, enabling efficient feature extraction. By using the MSDA, the model effectively reduces the impact of individual differences on experimental process, extracting domain-consistent features. Experiments evaluated on two EEG datasets demonstrated good performance in comparation with SOTA methods.
- 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 paper evaluated the method on two EEG datasets, including 1-2 more datasets would make the evaluation more solid.
- 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.
- 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/2025/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?
The presentation of the paper is good. The method has demonstrated superior performance in comparison with SOTA methods.
- Reviewer confidence
Somewhat confident (2)
- [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 appreciate the reviewers for their constructive comments, which have helped us refine our work further. Based on comments, we response to them here and will revise some parts of the manuscript. We acknowledge the widely discussed issue regarding reproducibility and open-source availability. We commit to organizing and releasing it on GitHub immediately upon the paper’s acceptance. @R1-About the issues of disorganized symbol management and multiple typos, we sincerely apologize for the compromised readability. We will meticulously revise the paper’s wording, details, and presentation in accordance with your suggestions, and restructure the paper to better clarify the design logic, rather than merely focusing on individually introducing each module of the structural design. @R1-About the main contributions of this paper. There are two prime contributions. 1), we design a novel architecture. Based on [17], this architecture integrating GCN, Spectral-Spatial Attention Mechanisms, and MSDA. The design of this structure aims to address the limitations of individual modules, the overfitting and transfer difficulty caused by the significant impact of individual differences when GCN extracts features, the insufficient spatial information extraction due to the non-Euclidean nature of EEG signals in SSAFE and the challenges MSDA faces in aligning overly complex features. Fused features preserve information richness and low redundancy, so this architecture can effectively extract domain-invariant features; 2), We design the MGAFE Module, which use dual-graph approach to extract features from different local spatial regions to addresses the issue of insufficient feature extraction from EEG signals using GCN combined with SSAFE. Based on this two contributes, VAG-MSDA can extract domain-consistent features in spectral domain, local and global spatial domain with low influence of individual differences. @R1-About the differ from reference [17]. This study primarily focuses on how to better construct a feature extraction network and reduce conflicts between the feature extraction network and MSDA. In contrast, reference [17] mainly focus on improving the MSDA method and does not emphasize the importance of the feature extraction network. This is the key difference between this paper and reference [17]. @R1-About the use of simple one-layer GCN and the its contribution analysis. Our ablation studies are to demonstrate the GCN combine with MSDA can effectively extract features and the architecture w/o MGAFE is better than GCN-MSDA and SSAFE-MSDA. Simple one-layer GCN serves as a coarse feature extractor in this architecture, which can ensure the lightweight of the network. It constitutes part of the network, and its effectiveness would be compromised without coordination with other modules. We ignore providing detailed textual explanations about its contribution analysis. If accepted, we will refine our presentation. @R3-About the omission of [1]. Reference [17] in our paper is the reference [1] and the SSAFE-MSDA is same as S2A2-MSDA. The article exists terminology inconsistency issues. We will rectify it. @R3-About the absence of a comprehensive surveys of MSDA and attention methods. We sincerely apologize for this neglect. We will Revise and augment other relevant MSDA and attention methods in revisions such as the [2]-[4]. @R3-About the concern about the lack of novelty in this paper. We acknowledge that most components of the proposed network structure are not entirely novel when examined individually. The primary focus of this work lies in the effective design of the overall architecture. Due to space constraints, this paper primarily focused on elaborating individual components of the network architecture for reproducibility while insufficiently addressing the overall design rationale and significance, which consequently led to this issue. If accepted, we will refine the descriptive details to make our intentions clearer.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
Meta-review #3
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Reject
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
The manuscript was rejected after rebuttal due to significant clarity issues, including typos, making it difficult to identify key contributions. The novelty was limited, as the proposed network structure heavily borrowed from existing works, and the integration of attention mechanisms and multi-source domain adaptation had been previously studied. Additionally, the design of attention modules, such as SSAFE and MGAFE, closely resembled established methods without adequate discussion or citation. These factors led to the conclusion that the manuscript did not meet publication standards.