List of Papers Browse by Subject Areas Author List
Abstract
Spiking Neural Networks (SNNs) enable energy-efficient, interpretable EEG recognition via sparse binary spikes. However, extracting discriminative features through spike-driven computations remains a bottleneck. Existing methods often resort to hybrid ANN-SNN architectures relying on floating-point matrix multiplications, which compromises SNN efficiency and obscures the transparency of spiking dynamics. To address this, we propose SpikeEEGformer, a spike-driven Transformer engineered to strengthen feature extraction from spike sequences. It integrates a Spike EEG Self-Attention for synergistic channel-temporal modeling and a Spatio-Temporal Spike Block to bridge adjacent spiking dependencies. Furthermore, we introduce Latent Token Spike Attention, a multi-timestep spike-friendly mechanism that stabilizes representation learning and improves generalization. Evaluated on five benchmarks, SpikeEEGformer consistently achieves state-of-the-art accuracy with competitive energy efficiency. Critically, analysis of spiking dynamics provides intrinsic interpretability by elucidating the decision-making process via distinctive firing behaviors, while saliency maps effectively identify the critical decision-driving channels.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2498_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{LuoJia_SpikeEEGformer_MICCAI2026,
author = { Luo, Jiacheng AND Sun, Shuifa AND Tao, Qin AND Cui, Wenchao AND Wang, Ben},
title = { { SpikeEEGformer: A Spike-Driven Transformer for Generalized EEG Recognition } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16894},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper introduces SpikeEEGformer, a new spike-based Transformer model for EEG recognition that maintains the event-driven properties of spiking neural networks (SNNs). Its main contribution is the removal of floating-point matrix multiplications by implementing a spike-compatible attention mechanism.
- 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 SESA mechanism offers a spike-driven alternative to traditional attention, eliminating expensive floating-point calculations and lowering complexity from quadratic to linear (O(C + T)). This represents a significant advancement in neuromorphic AI.
- 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 baseline comparison is limited; the paper lacks the benchmarking against recent EEG Transformers. 2.The datasets are clinically relevant, but there is no real world deployment. 3.The paper doesn’t state anything about the code used. 4.Ablation depth must be improved.
- 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
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 paper offers a valuable and technically robust contribution to spike-based EEG modeling, introducing a novel attention mechanism and solid empirical validation. The methodological innovations, especially the spike-driven attention design and efficiency enhancements, are noteworthy and highly relevant to the MICCAI community. However, the work falls just short of a strong accept because it offers limited comparison with the latest models, lacks clinical validation, and has minor clarity issues. With enhancements in positioning and reproducibility, it could have a substantial impact.
- 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
The paper proposes a set of novel blocks for energy efficient EEG signal processing. The new blocks are derived from spiking neural networks and aim to eliminate the need for energy and computationally expensive floating point operation for simpler operations like additions or binary shifts. The paper is well validated on several standard datasets and compared to recent techniques achieves improved performance across them all despite reduced energy needs.
- 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 novel attention modules for EEG signal processing architectures using lower energy / non-floating point operations.
Strong benchmarking of current techniques and reference datasets and comparison with the proposed approach showing significant performance gains on reference datasets.
While the architecture is not the most energy efficient among the evaluated techniques, the proposed SpikeEEGformer-S strikes a good balances between performance (near SOTA) and energy use.
- 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.
Evaluation of the method discrete predictions (e.g. binary valence in DEAP) only. How well does this method perform for regressive tasks?
The reproducibility of the paper can be a bit difficult and the clarity could be slightly improved.
- What are parameters of the Spike encoder?
- The STSB block on Fig. 2 seems to imply that 2-3 is treated differently from 1-2 and 2-3. - How are saliency maps computed?
- Can you detail the conversion from T timestamps to T_s? Are these uniform timestamps or only when the spikes happen?
- 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?
The paper is enjoyable to read, pushes the performance boundary, has considerations for energy usage of machine learning models, the proposed components can should be reused in other applications where with binary inputs to transformer-like methods are applicable.
- 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
Review #3
- Please describe the contribution of the paper
This paper proposes SpikeEEGformer, a full spike-driven Transformer architecture for EEG recognition, designed to address limitations of existing Spiking Neural Network (SNN) models, particularly their reliance on hybrid ANN-SNN computations and difficulty in extracting discriminative features from spike trains. The method introduces three main components : 1.Spike EEG Self-Attention (SESA): a spike-compatible attention mechanism designed to model channel-temporal dependencies without relying on floating-point matrix multiplications (Section 2.4). 2.Spatio-Temporal Spike Block (STSB): a dual-branch module combining temporal spike modeling and spatial feature extraction to mitigate information decay across spiking steps (Section 2.5). 3.Latent Token Spike Attention (LTSA): module that stabilizes representation learning by decoupling attention from input noise (Section 2.6). The model is evaluated on five EEG benchmarks across psychiatric dignosis, emotion recognition, and motor imagery tasks. The proposed method achieves state-of-the-art performance while maintaining competitive energy efficiency compared to ANN and hybrid SNN baselines.
- 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 paper addresses an important and well-motivated problem in EEG analysis. The limitations of hybrid ANN-SNN approaches and the difficulty of extracting meaningful representations from spike trains are clearly identified in the introduction (page 2). One key strength of the paper is the design of a fully spike-driven Transformer architecture, avoiding floating-point matrix multiplications. The effort to maintain spike-based computation throughout the model is well aligned with the objectives of SNNs. For example, SESA replaces conventional attention operations with additive and spike-based computations (Section 2.4), and STSB uses hardware-friendly operations such as bit-shift scaling (Section 2.5). The architecture is modular, coherent, and well structured : the overall pipeline (Fig. 1, page 3) clearly illustrates a hierarchical design with progressive spatio-temporal abstraction. The different components (SESA, STSB, LTSA) are logically motivated and target complementary limitations of spike-based learning, including temporal dependency modeling and training stability. Another strong point is the relatively comprehensive experimental evaluation. The method is validated on five datasets across different EEG paradigms (Section 3.1), which is stronger than many works focusing on a single task. The paper also includes ablation studies (Table 4) that analyze the contribution of each module in the full pipeline. Finally, the paper makes an effort toward model interpretability by analyzing spike firing patterns and channel saliency maps (Section 3.5-3.6, Fig. 3), which is an interesting direction in the context of SNN-based models
- 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 experimental results require a more careful interpretation. As shown in Table 1, the reported performance improvements over strong SNN/ANN baselines are relatively modest compared to the standard deviations (e.g., 93.89 + 6.62 vs 91.12 + 7.73), which raises questions about the statistical significance and robustness of the gains. In addition, the claims regarding energy efficiency are not fully convincing. While SNNs are generally expected to be more energy-efficient than ANN-based models, this advantage is not consistently reflected in Table 1, where some ANN baselines achieve lower energy consumption than several SNN-based approaches. The interpretability analysis remains limited. Although the paper presents spike firing statistics and saliency maps (Fig 3), these analyses are mostly qualitative and lack rigorous validation. There is no quantitative evaluation, nor comparison with established explainability methods.
- 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
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 paper proposes a well-structured and technically sound approach for EEG recognition using a fully spike-driven Transformer architecture. The idea of preserving spike-based computation while improving feature extraction is well motivated, especially in the context of energy-efficient and neuromorphic systems. The overall design is coherent, and the evaluation on multiple datasets is a strong point of the paper. However, some aspects limit the overall impact of the work. First, the reported improvements over strong baselines are relatively small compared to the standard deviations (Table 1), which makes it difficult to clearly assess the robustness of the gains. Second, although energy efficiency is presented as an important motivation, the results do not always show a clear advantage over ANN-based models, and the analysis is based on theoretical estimations rather than real measurements. Finally, the interpretability part is interesting but remains quite limited, as it relies mainly on qualitative observations without deeper validation. Overall, I think the paper is solid and presents interesting ideas, but some claims would benefit from a more rigorous analysis. In particular, the paper could be improved by: 1.Providing a more detailed analysis of performance variability, 2.Strengthening the evaluation of energy efficiency, 3.Including a more thorough assessment of the interpretability component.
- 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.
The paper presents a technically sound and well-motivated methodological contribution, with consistent agreement across reviewers on its novelty, solid design, and comprehensive multi-dataset evaluation. Despite some concerns regarding benchmarking depth and clarity, the identified issues are incremental and do not undermine the core contribution, which is sufficiently strong to justify an early accept.
No rebuttal is strictly necessary; however, if provided, clarifications on baseline comparisons, energy efficiency analysis, and implementation details would further strengthen the final version.