The DOI matches its Crossref record; 48 of the 48 references it cites were found in a registry or an index; 7 of the 20 sources its claims cite were checked for support.
ST-CapsNet: Linking Spatial and Temporal Attention With Capsule Network for P300 Detection Improvement
Open accessAbstract
A brain-computer interface (BCI), which provides an advanced direct human-machine interaction, has gained substantial research interest in the last decade for its great potential in various applications including rehabilitation and communication. Among them, the P300-based BCI speller is a typical application that is capable of identifying the expected stimulated characters. However, the applicability of the P300 speller is hampered for the low recognition rate partially attributed to the complex spatio-temporal characteristics of the EEG signals. Here, we developed a deep-learning analysis framework named ST-CapsNet to overcome the challenges regarding better P300 detection using a capsule network with both spatial and temporal attention modules. Specifically, we first employed spatial and temporal attention modules to obtain refined EEG signals by capturing event-related information. Then the obtained signals were fed into the capsule network for discriminative feature extraction and P300 det- ection. In order to quantitatively assess the performance of the proposed ST-CapsNet, two publicly-available datasets (i.e., Dataset IIb of BCI Competition 2003 and Dataset II of BCI Competition III) were applied. A new metric of averaged symbols under repetitions (ASUR) was adopted to evaluate the cumulative effect of symbol recognition under different repetitions. In comparison with several widely-used methods (i.e., LDA, ERP-CapsNet, CNN, MCNN, SWFP, and MsCNN-TL-ESVM), the proposed ST-CapsNet framework significantly outperformed the state-of-the-art methods in terms of ASUR. More interestingly, the absolute values of the spatial filters learned by ST-CapsNet are higher in the parietal lobe and occipital region, which is consistent with the generation mechanism of P300.
Overview
CoChat Summary
TL;DR
ST-CapsNet combines spatial and temporal attention modules with a capsule network to refine EEG signals before P300 detection. Across BCI Competition datasets, it improved single-trial P300 detection and cumulative symbol recognition over ERP-CapsNet and several CNN/traditional baselines.
Key findings
- Single-trial P300 detection: ST-CapsNet reached Acc./F1 of 0.8997/0.7213 on IIb, 0.7785/0.4535 on II-A, and 0.8342/0.5405 on II-B; ERP-CapsNet reached 0.8927/0.7146, 0.7456/0.4546, and 0.8160/0.5287, respectively. ST-CapsNet outperformed other models on IIb and II-B, while ERP-CapsNet had slightly higher F1 on II-A.
- Symbol recognition: ST-CapsNet had the highest ASUR15 on all three datasets (IIb 30.5, II-A 74.0, II-B 84.3), versus ERP-CapsNet’s 30.1, 73.7, and 82.9. ERP-CapsNet was slightly more accurate in the first 5 repetitions on II-A (ASUR5 45.8 vs. 45.2).
- ITR: ST-CapsNet and MsCNN-TL-SVM achieved the highest ITR on IIb (51.56 bits/min); ST-CapsNet reached 13.32 bits/min in the 6th repetition on II-A and 19.74 bits/min in the 2nd repetition on II-B.
- Learned spatial attention maps shared channels Cz and CPz and highlighted central/parietal regions; averaged spatial filters had higher values in parietal and occipital regions, with common top electrodes PO7, PO8, O1, CPz, and Pz.
Methods
- Design: EEG signals were first refined by spatial attention and temporal attention modules, then fed into a capsule network for discriminative feature extraction and P300 classification. The capsule network used 10 C×1 spatial filters, 64 1×13 temporal filters, 48 8D primary capsules, 3 dynamic-routing iterations, two 16D output capsules, and a decoder with 512, 1024, and C×78 neurons.
- Data: Two public datasets were used—Dataset IIb of BCI Competition 2003 and Dataset II of BCI Competition III, with the latter split into II-A and II-B. Preprocessing extracted 0–650 ms after intensification, applied a 0.1–20 Hz FIR band-pass filter, downsampled to 78 time points, and normalized via z-score and sigmoid; P300/Non-P300 imbalance was reduced by averaging randomly selected P300 samples.
- Analysis: Models were implemented in PyTorch, trained on an RTX 2080 Ti with batch size 64, initial learning rate 0.001 with exponential decay 0.96, Adam optimizer, early stopping, and braindecode augmentation. Evaluation used accuracy and F1 for single-trial P300 detection, the proposed ASUR metric for cumulative symbol recognition across repetitions, and ITR.
Limitations
- Stated: the capsule network has relatively many parameters compared with traditional methods and CNNs, requiring longer training time and higher-performance equipment.
- Stated: the model cannot precisely control recognition accuracy at a single repetition because P300 detection in single trial and symbol recognition are separate tasks, and the model/loss were designed for the first task without a dedicated training method for symbol recognition.
- Evident from the text: evaluation is limited to two public P300 datasets, split into three subject-specific datasets (IIb, II-A, II-B), so generalization to other subjects, paradigms, or larger cohorts is not established.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
7 Valid · 1 Not valid · 7 Not checked · 50 Own results and interpretations
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“the learned spatial attention maps generally accords with those of previous studies [15], [23], [24].”
What the sources say
- Not checked — no full text of the sourcecites [15] source
- Supportedcites [23] source
Section 5.2 (page unavailable) describes spatial maps learned from CNN weights and identifies discriminant electrodes around Pz, Cz, PO7, and PO8. The paper therefore provides a previous learned spatial pattern of the kind the citing sentence says its maps generally accord with. (LLM, conf=0.86)
- Not checked — no full text of the sourcecites [24] source
hedgedstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“However, ERP-CapsNet just took the raw EEG signals as input, which introduced additional noise.”
What the sources say
- Partly supportedcites [29] sourceExperimental — accuracy not yet measured
The paper describes ERP-CapsNet as detecting ERPs from raw EEG signals, but its Data Preprocessing section (page unavailable) says the signals were filtered, downsampled, and normalized before entering the network. The paper also says ERP signals are embedded in background noise; it does not say that using EEG as input introduced additional noise. (LLM, conf=0.86)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The P300 potentials are different from the NonP300 potentials, because the rare target stimuli cause subjects' brains to generate P300 potentials [33].”
What the sources say
- Partly supportedcites [33] sourceExperimental — accuracy not yet measured
The Introduction (p. 1) describes the P300 component in a spelling system that detects potentials evoked by flashing rows and columns. It does not state that rare target stimuli cause P300 potentials or explicitly contrast P300 with Non-P300 potentials. (LLM, conf=0.83)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“More interestingly, the absolute values of the spatial filters learned by ST-CapsNet are higher in the parietal lobe and occipital region, which is consistent with the generation mechanism of P300.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“our ST-CapsNet outperforms ERP-CapsNet by about 1 percent and is better than the other models in symbol recognition.”
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
37 more claims
Create a free account to see all 65 claims and to verify the 1 reference still open.
- Probably not a substantial claim. 12
- It may state two separate claims as one. 13
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“ST-CapsNet is an enhanced version of ERP-CapsNet[45] that combined spatial-temporal attention and Capsule Network for Improved P300 detection[40].”
From A Cascade xDAWN EEGNet Structure for Unified Visual-Evoked Related Potential Detection (2024) · cites this paper as [40]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ C. Capsule Network · p. 4
“Next, we used 64 1 × 13 temporal filters by convolution and ReLU operations to extract temporal features of which size is 64 × 1 × 8.”
§ C. Capsule Network · p. 5
“| Covn2D | 64 | (1, 13) (64, 1, 6) | ReLU | 13 |”
Citations
Cites
Squeeze-and-Excitation Networks
Hu, J., Shen, L., & Sun, G. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00745 · Cited by 35,667 · cited 1×
“Several attention mechanisms have been widely used, such as the Squeeze-Excitation (SE) block [30] proposed by Hu et al. which adaptively generates channel attention maps and recalibrates the feature responses of channels by explicitly modelling the interdependencies between channels.”
MethodCBAM: Convolutional Block Attention Module
Woo, S., Park, J., Lee, J.-Y., & Kweon, I. S. (2018) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-01234-2_1 · Cited by 25,853
Brain–computer interfaces for communication and control
Wolpaw, J. R., Birbaumer, N., McFarland, D. J., Pfurtscheller, G., & Vaughan, T. M. (2002) · Clinical Neurophysiology · doi:10.1016/s1388-2457(02)00057-3 · Cited by 7,927 · cited 1×
“These datasets are complete records of P300 evoked potentials recorded with BCI2000 [32] using a paradigm described by Farwell and Donchin [14].”
MethodCBAM: Convolutional Block Attention Module
Woo, S., Park, J., Lee, J.-Y., & Kweon, I. S. (2018) · arXiv · doi:10.48550/arxiv.1807.06521 · Cited by 6,261 · cited 2×
“We set r to 16 as suggested in [31].”
BackgroundMethod(2005) · Cited by 5,977
EEGNet: a compact convolutional neural network for EEG-based brain-computer interfaces
Lawhern, V. J., Solon, A. J., Waytowich, N. R., Gordon, S. M., Hung, C. P., & Lance, B. J. (2018) · Journal of Neural Engineering · doi:10.1088/1741-2552/aace8c · Cited by 5,115 · cited 1×
“Zhang et al. proposed an improved EEGNet [19] that combined xDAWN saptial filtering with EEGNet [20] for the individually-calibrated rapid serial visual presentation (RSVP) task and won second place in the BCI Controlled Robot Contest at 2022 World Robot Con-test [21].”
MethodEEGNet: A Compact Convolutional Network for EEG-based Brain-Computer Interfaces
Lawhern, V. J., Solon, A. J., Waytowich, N. R., Gordon, S. M., Hung, C. P., & Lance, B. J. (2016) · arXiv · doi:10.48550/arxiv.1611.08024 · Cited by 5,105
Dynamic Routing Between Capsules
Sabour, S., Frosst, N., & Hinton, G. E. (2017) · arXiv · doi:10.48550/arxiv.1710.09829 · Cited by 4,167
MEG and EEG data analysis with MNE-Python
Gramfort, A. (2013) · Frontiers in Neuroscience · doi:10.3389/fnins.2013.00267 · Cited by 3,784 · cited 1×
“Next, we used MNE-Python [45] to plot the topography of datasets IIb, II-A and II-B.”
MethodDeep learning with convolutional neural networks for EEG decoding and visualization
Schirrmeister, R. T., Springenberg, J. T., Fiederer, L. D. J., Glasstetter, M., Eggensperger, K., Tangermann, M., Hutter, F., Burgard, W., & Ball, T. (2017) · Human Brain Mapping · doi:10.1002/hbm.23730 · Cited by 3,751
Research works printed but not linked: 2
- Wang, 2019, An event related potential electroencephalogram signal analysis method based on denoising auto-encoder neural network, Control Theory Appl., 36, 589
- Christof, 2021, Lossfunctions.jl.
Cited by
Driving Fatigue Detection Based on Hybrid Electroencephalography and Eye Tracking
Lian, Z., Xu, T., Yuan, Z., Li, J., Thakor, N. V., & Wang, H. (2024) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2024.3446952 · Cited by 89
“It’s important to note that the weights learned for 1D features may have reduced interpretability compared to those applied to earlier-stage features, such as temporal attention (used for time samples) or spatial attention (used for channels) [55].”
BackgroundSelf-citationWang, H., Yuan, Z., Zhang, H., Wan, F., Li, Y., & Xu, T. (2025) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2025.107570 · Cited by 40
Self-citationConvolutional Dynamically Convergent Differential Neural Network for Brain Signal Classification
Zhang, Z., He, Y. L., Mai, W., Luo, Y., Li, X., Cheng, Y., Huang, X., & Lin, R. (2024) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2024.3437676 · Cited by 38
“Subsequently, we calculated the accuracy of the proposed method on the test dataset and compared it with current methods [7], [8], [15], [16], [17], [20], [22], [23].”
EEG–fNIRS-Based Emotion Recognition Using Graph Convolution and Capsule Attention Network
Chen, G., Liu, Y., & Zhang, X. (2024) · Brain Sciences · doi:10.3390/brainsci14080820 · Cited by 36
“Third, fNIRS signals can be converted to concentration changes of HbO and HbR.”
BackgroundImproved dilation CapsuleNet for motor imagery and mental arithmetic classification based on fNIRS
Li, Y., Xu, T., Li, J., Wan, F., & Wang, H. (2024) · Brain-Apparatus Communication A Journal of Bacomics · doi:10.1080/27706710.2024.2335886 · Cited by 25
Self-citation
Metadata
- Authors
- Wang, Z., Chen, C., Li, J., Wan, F., Sun, Y., & Wang, H.
- Affiliations
- Wuyi University (CN)
- University of Essex (GB)
- University of Macau (MO)
- Sir Run Run Shaw Hospital (CN)
- Venue
- IEEE Transactions on Neural Systems and Rehabilitation Engineering
- Published
- 2023
- Volume
- 31
- Pages
- 991-1000
- ISSN
- 1534-4320 (ISSN-L)
- 1558-0210
- OpenAlex
- W4316660615
- Cited by
- 75
- References
- 43
- Citations by year
- 2026: 19 · 2025: 23 · 2024: 29 · 2023: 4
- FWCI
- 11.25
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Locations
- IEEE Transactions on Neural Systems and Rehabilitation Engineering (publishedVersion, cc-by)
- PubMed (publishedVersion)
- Open Access at Essex (University of Essex) (acceptedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- Special Projects in Key Fields Supported by the Technology Development Project of Guangdong (2020ZDZX3018)
- Special Fund for Science and Technology of Guangdong Province (2020182)
- Wuyi University and Hong Kong and Macao joint Research Project (2019WGALH16)
- Basic and Applied Basic Research Foundation of Guangdong Province (2020A1515111154)
- Department of Education of Guangdong Province (2021KTSCX136)
- National Natural Science Foundation of China (82172056)
- Zhejiang Provincial Natural Science Foundation of China (LR23F010003)
- National Key Research and Development Program of China (2021ZD030048)
- Key Research and Development Program of Zhejiang Province (2022C03064)
- Science and Technology Development Fund, Macau (0045/2019/AFJ)
- University of Macau (MYRG2022-00197-FST)
- Topics
- Life Sciences › Neuroscience › Cognitive Neuroscience › EEG and Brain-Computer Interfaces
- Physical Sciences › Computer Science › Human-Computer Interaction › Gaze Tracking and Assistive Technology
- Physical Sciences › Computer Science › Artificial Intelligence › Emotion and Mood Recognition
- Keywords
- P300 detection, brain-computer interface, capsule networks, spatial attention, temporal attention, symbol recognition, BCI competition datasets
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: ST-CapsNet: Linking Spatial and Temporal Attention With Capsule Network for P300 Detection Improvement. https://app.cochat.ai/papers/10.1109/tnsre.2023.3237319
- The paper
- Wang, Z., Chen, C., Li, J., Wan, F., Sun, Y., & Wang, H. (2023). ST-CapsNet: Linking Spatial and Temporal Attention With Capsule Network for P300 Detection Improvement. IEEE Transactions on Neural Systems and Rehabilitation Engineering. https://doi.org/10.1109/tnsre.2023.3237319
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.