Pith. sign in

REVIEW 3 major objections 6 minor 32 references

Subject Specific Deep Learning Model for Motor Imagery Direction Decoding

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A subject-specific EEGNet variant with Squeeze-and-Excitation ranking layers reaches 58.77% average accuracy for online decoding of motor-imagery direction, beating EEGNet and FBCNet baselines.

desk verdict Modest, plausibly real gain for directional MI decoding, undermined by a self-identified ocular artifact confound and no significance testing. read the letter →

arxiv 2501.01725 v1 pith:JGXWNDVW submitted 2025-01-03 eess.SP cs.NE

Praveen K. Parashiva , Sagila Gangadaran , A. P. Vinod This is my paper · ORCID
classification eess.SPcs.NE
keywords motorimagerybrain-computerinterfacedirectiondecodingunilateralEEGSqueeze-and-Excitationelectroderankingfeature-map
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that adding Squeeze-and-Excitation (SE) ranking layers to an EEGNet-style convolutional network improves online decoding of left-versus-right direction from unilateral motor imagery in EEG, a task that matters because it could give stroke patients a more natural, higher-dimensional BCI control signal than standard left-hand/right-hand imagery. On data from 20 healthy subjects, with an online session for 13 of them, the paper reports 58.28 ± 9.00% average accuracy for the subject-independent model and 58.77 ± 8.10% for the highest-accuracy subject-specific fine-tuned variant, compared with 54.50 ± 5.75% for EEGNet and 49.99 ± 6.73% for FBCNet. The claimed gain comes from learned ranking layers that reweight electrodes and feature maps, and the same ranking outputs double as an interpretability map. The subject-specific model also shows that only the ranking and dense layers need fine-tuning, not the convolutional feature extractor.

What carries the argument

The central mechanism is the Squeeze-and-Excitation ranking layer applied along two axes. The electrode-ranking layer averages each channel across time, passes the resulting vector through a small dense bottleneck with ReLU and sigmoid activations, and multiplies each electrode's signal by its learned importance weight; the feature-map-ranking layer performs the same operation across the spatial and temporal dimensions of each convolutional filter. The paper fixes the EEGNet convolution filters after pre-training and updates only these ranking layers plus the dense layer for each subject, so the SE weights carry the subject-specific adaptation. The output weights are also the paper's interpretability result: they can be visualized as heatmaps and filter-rank plots to see which electrodes and filters drive the decision.

What would settle it

Run the same calibration and online protocol with horizontal and vertical EOG channels recorded and removed by regression or rejection, or with subjects instructed to keep gaze fixed during the MI cue, and compare the average accuracy. If the 58.8% advantage over EEGNet collapses toward chance or toward the baseline accuracy, the central claim fails.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that SE blocks can be moved from image-channel attention to EEG electrode and feature-map attention, and that this is enough to beat standard deep decoders on directional unilateral MI. The model pre-trains a subject-independent base on calibration data from subjects S01-S07, then for each online subject fine-tunes only the electrode-ranking layer, the feature-map-ranking layer, and the dense layer. The highest-accuracy subject-specific configuration reaches 58.77 ± 8.10% average online accuracy, with the top per-subject accuracy 70.31% (S14 and S20) and the lowest 43.75% (S10). The electrode heatmaps assign high importance to central electrodes near the motor cortex (FC3, FCz, FC4, C3, Cz, C2, CP1, CPz, CP4) and low importance to frontal electrodes such as Fp1, Fp2, AF3, and AFz, which the paper reads as evidence that the ranking is capturing motor-related cortical activity.

Load-bearing premise

The results stand on the assumption that the model is learning genuine motor-imagery direction information and not eye-movement artifacts: preprocessing includes bandpass, notch, baseline correction, and common average referencing but no EOG artifact rejection, and the paper itself attributes the high rank of electrode AF7 to noise and eye-related artifacts.

Editorial extensions

If this is right

  • If the reported accuracy is right, fine-tuning only the ranking and dense layers outperforms continued training of all weights, so the learned importance weights, not the convolutional filters, are what adapt the model to a new subject.
  • The high ranks assigned to motor-cortex electrodes and low ranks to prefrontal electrodes suggest the ranking layers could serve as a data-driven channel-selection map before a real deployment.
  • Because the feedback-period EEG was excluded, the online results are attributable to the 4-second motor-imagery window alone, which means the architecture itself, not feedback artifacts, carries the decoding.
  • For stroke rehabilitation, the framework offers a way to decode left/right intent from a single limb, which would give a BCI more degrees of freedom than conventional bilateral MI without requiring a new feature-engineering pipeline per patient.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The authors do not test artifact rejection, and they attribute the high rank of electrode AF7 to eye-related noise; an immediate extension would be to compare accuracy with and without EOG regression, which would isolate how much of the 58.8% is neural direction information.
  • The same SE-ranking recipe could be dropped into other EEG-BCI paradigms with shift-invariant channel relevance, such as P300 spellers or SSVEP control, where per-user electrode importance could be learned in a few trials.
  • Their subject-independent model already sits at 58.28%, so a zero-calibration deployment may be feasible; a follow-up needs to check whether the extra 0.5% from fine-tuning justifies the 10-minute calibration session in practice.
  • A natural next comparison is to run the same subject-independent/subject-specific split against the self-attention unilateral-MI decoder reported in the introduction, since its reported roughly 59% accuracy is the closest point of reference for the proposed method.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. This paper proposes a deep learning architecture for online decoding of left/right directional information from unilateral motor imagery in 20 healthy subjects. The architecture augments EEGNet with Squeeze-and-Excitation (SE) layers that rescale electrodes and feature maps, and the authors evaluate a subject-independent base model and several subject-specific fine-tuned variants on held-out online sessions (13 subjects, 48 trials per subject). The central claim is that the best subject-specific configuration reaches 58.77 ± 8.10% average online accuracy, outperforming EEGNet and FBCNet. The paper also analyzes electrode and filter rankings as an interpretability result.

Significance. The question of decoding directional information from unilateral MI in an online setting is relevant to natural BCI control, and the use of held-out online session data is a strength: the reported means are direct online measurements rather than offline cross-validation. The SE-based ranking mechanism is a simple, transferable idea, and the electrode/filter-rank visualizations offer a useful interpretability tool. However, the central outperformance claim is not yet statistically supported, and the lack of ocular artifact control leaves an unaddressed confound that could explain part of the advantage. If the necessary robustness and significance analyses confirm the effect, the contribution would be a modest but useful step for online directional MI decoding.

major comments (3)
  1. [Section III.A, Tables I and II] The claim that the proposed method outperforms EEGNet and FBCNet is based on group means without any significance testing. With n = 13 subjects, the between-model gaps are within the across-subject variability: for the subject-independent comparison (58.28 ± 9.00 vs. 54.50 ± 5.75), a paired t-test would require per-subject differences, but the reported SDs imply an approximate t-statistic below 2, which is not significant at the 5% level. The subject-specific comparison (58.77 ± 8.10 vs. 55.28 ± 10.05) is likewise inconclusive. Please report per-subject accuracies, paired confidence intervals, and p-values, and state the decision rule for claiming superiority.
  2. [Section II.B and Section III.B] The preprocessing (0.5–90 Hz bandpass, notch, CAR) does not include EOG rejection, and Section III.B explicitly attributes the high rank of electrode AF7 to noise and eye-related artifacts. Since the electrode ranking layer (Eq. 1a–1c) scales whole channels based on the trial-mean voltage, a low-frequency ocular offset that differs between left and right cues can directly boost classification. The observed ~4-point advantage over EEGNet is therefore potentially artifact classification rather than neural direction decoding. The authors should test whether the result survives removal of frontal/ocular components (e.g., ICA or regression) or provide concurrent EOG evidence.
  3. [Section III.A, Fig. 4] With only 48 online trials per subject, the standard error of a single-subject accuracy is about 7.2 percentage points. The observed range (e.g., S10 at 43.75%, S14/S20 at 70.31%) includes many values that are not individually distinguishable from chance. The paper should report binomial chance-level tests per subject and either exclude or separately analyze participants who perform at or below chance, because averaging over such subjects can bias the headlined mean and the model-comparison conclusions.
minor comments (6)
  1. [Equation (2a)] The pooled feature vector in (2a) uses X(i, j) although the input to the feature-map ranking layer is Y; it should be Y(i, j).
  2. [Fig. 2] Several labels in the architecture diagram are incomplete (e.g., 'on D', ' atch ormali ation'), making the figure difficult to interpret.
  3. [Abstract vs. Table II] The abstract reports 58.7 ± 8% while Table II reports 58.77 ± 8.10%; please make these values consistent.
  4. [Section III.C] The filter-ranking observations are based on only three subjects and are post hoc; the text should label them as exploratory rather than as confirmatory evidence.
  5. [Reference [18]] Reference [18] (EEGNet) is missing its full citation details, including the journal or conference and year.
  6. [Table II] The term 'Continued Training' is not defined in the text; please specify whether it fine-tunes all layers and with what learning-rate schedule.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported accuracies are held-out online-session results, and the SE-layer ranks are learned weights rather than fitted constants presented as predictions.

full rationale

The paper's central quantities are online direction-decoding accuracies. The proposed model is trained on calibration-session data (S01-S07 for the base model; per-subject calibration data for fine-tuned variants) and evaluated on a separate online session (S08-S20), so the reported accuracies are held-out test results rather than quantities reconstructed from the training fit. Equations (1a)-(1c) and (2a)-(2c) define electrode and feature-map scaling factors that are learned model parameters optimized with cross-entropy on calibration labels; they do not encode or reproduce the test accuracies. Baseline comparisons to EEGNet, FBCNet, W-CSP, and W-CSP&W-PLV are external methods evaluated on the same held-out online data, not outputs of the proposed model. Self-citations [11] and [30] provide prior experimental details and comparison baselines, but they do not carry the central accuracy claim. The electrode and feature-map ranking analyses in Sections III.B-C are post-hoc interpretations of learned weights and do not feed back into the accuracy calculation. The derivation chain is therefore self-contained, and no fitted parameter is renamed as a prediction. Concerns about ocular artifacts would bear on external validity and artifact control, not on circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim is empirical, so the main load-bearing items are dataset and artifact assumptions plus architecture and hyperparameter choices, rather than mathematical postulates. No new physical entities are introduced. The most fragile assumption is that residual eye artifacts do not carry direction information; the paper itself flags AF7 noise. Hyperparameters and preprocessing choices are numerous and may not generalize to other datasets.

free parameters (4)
  • SE reduction rate r = 3
    Selected by hand for all subjects; controls the width of the bottleneck in the electrode and feature-map ranking layers (Sections II.C and III).
  • Convolution filter counts = 8, 16, 16
    Architecture choice for the three convolution layers; the feature-map ranking analysis in Section III.C depends on these specific filter counts.
  • Training hyperparameters = learning rate 1e-4, batch size 32, max 1000 epochs, patience 30, min delta 1e-3, min epochs 100
    Reported in Section III as fixed choices; they influence the achieved accuracy and early stopping behavior.
  • Preprocessing parameters = 0.5-90 Hz 5th-order Butterworth, 50 Hz notch, CAR, 4 s MI-cue window
    Chosen in Section II.B; segmentation and filtering choices determine what signal the model sees and whether artifacts remain.
assumptions (5)
  • domain assumption The 4 s MI cue EEG, after bandpass, notch, baseline correction, and CAR, contains decodable directional information for left versus right unilateral motor imagery.
    Invoked in Sections II.A and II.B; if this fails, the accuracy numbers are not measuring the intended neural signal.
  • domain assumption Same-day calibration data are representative of the online session EEG for each subject.
    The base model is trained or fine-tuned on calibration data and tested on online session data (Section II.A); session drift would invalidate the comparison.
  • domain assumption Residual eye-movement and ocular artifacts do not systematically encode the imagined direction.
    No EOG rejection is applied, and Section III.B attributes a high AF7 rank to noise and eye-related artifacts; direction-correlated artifacts would confound the decoder.
  • domain assumption Convolution filters learned from calibration data of subjects S01-S07 remain useful fixed feature extractors for subjects S08-S20.
    Fine-tuning in Section III.A updates only the ranking and dense layers while keeping convolution weights fixed; poor transfer would break the subject-specific result.
  • standard math Standard deep learning and signal-processing background, including convolution, backpropagation, Adam, and the EEGNet architecture, behaves as assumed.
    The model relies on standard implementations of EEGNet and SE blocks; the paper provides no formal proof of convergence or optimality.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Subject Specific Deep Learning Model for Motor Imagery Direction Decoding." pith.science (2026). https://pith.science/paper/JGXWNDVW

@misc{pith2026250101725,
  author       = {Pith},
  title        = {Pith review of: Subject Specific Deep Learning Model for Motor Imagery Direction Decoding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JGXWNDVW}},
  note         = {Machine review of arXiv:2501.01725}
}
read the original abstract

Hemispheric strokes impair motor control in contralateral body parts, necessitating effective rehabilitation strategies. Motor Imagery-based Brain-Computer Interfaces (MI-BCIs) promote neuroplasticity, aiding the recovery of motor functions. While deep learning has shown promise in decoding MI actions for stroke rehabilitation, existing studies largely focus on bilateral MI actions and are limited to offline evaluations. Decoding directional information from unilateral MI, however, offers a more natural control interface with greater degrees of freedom but remains challenging due to spatially overlapping neural activity. This work proposes a novel deep learning framework for online decoding of binary directional MI signals from the dominant hand of 20 healthy subjects. The proposed method employs EEGNet-based convolutional filters to extract temporal and spatial features. The EEGNet model is enhanced by Squeeze-and-Excitation (SE) layers that rank the electrode importance and feature maps. A subject-independent model is initially trained using calibration data from multiple subjects and fine-tuned for subject-specific adaptation. The performance of the proposed method is evaluated using subject-specific online session data. The proposed method achieved an average right vs left binary direction decoding accuracy of 58.7 +\- 8% for unilateral MI tasks, outperforming the existing deep learning models. Additionally, the SE-layer ranking offers insights into electrode contribution, enabling potential subject-specific BCI optimization. The findings highlight the efficacy of the proposed method in advancing MI-BCI applications for a more natural and effective control of BCI systems.

Figures

Figures reproduced from arXiv: 2501.01725 by the authors.

Figure 1
Figure 1. a) Electrode Placement and b) Timing Diagram [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 4
Figure 4. Comparison of Online Unilateral MI Direction Decoding Accuracy [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Heatmap to illustrate the electrode ranking layer on subjects S08 [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figures from the paper (1 more)
Figure 6
Figure 6. Figure 6: Filter rank assigned at A) Layer 1, B) Layer 2, and C) Layer 3 [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 17 canonical work pages

  1. [1]

    rain - computer interfaces in neurorecovery and neurorehabilitation,

    M. J. Young, D. J. Lin, and L. . Hochberg, “ rain - computer interfaces in neurorecovery and neurorehabilitation,” Semin Neurol , vol. 41, no. 2, p. 206, Apr. 2021, doi: 10.1055/S-0041-1725137

    work page doi:10.1055/s-0041-1725137 2021

  2. [2]

    rain–Computer Interfaces,

    He, H. Yuan, J. Meng, and S. ao, “ rain–Computer Interfaces,” Neural Engineering: Third Edition , pp. 131–183, Jan. 2020

    work page 2020

  3. [3]

    Filter bank common spatial pattern algorithm on I competition IV datasets a and b,

    K. K. Ang, Z. Y. Chin, C. Wang, C. Guan, and H. Zhang, “Filter bank common spatial pattern algorithm on I competition IV datasets a and b,” Front Neurosci, vol. 6, no. MAR, 2012, doi: 10.3389/fnins.2012.00039

    arXiv 2012

  4. [4]

    on -invasive Brain -Computer Interfaces State of the rt and Trends,

    B. J. Edelman et al. , “ on -invasive Brain -Computer Interfaces State of the rt and Trends,” IEEE Rev Biomed Eng , pp. 1 –25, 2024, doi: 10.1109/RBME.2024.3449790

    work page arXiv 2024

  5. [5]

    From classic motor imagery to comple mo ement intention decoding: The noninvasive Graz -BCI approach,

    G. R. Müller-Putz, A. Schwarz, J. Pereira, and P. Ofner, “From classic motor imagery to comple mo ement intention decoding: The noninvasive Graz -BCI approach,” Prog Brain Res , vol. 228, pp. 39 –70, 2016, doi: 10.1016/bs.pbr.2016.04.017

    work page doi:10.1016/bs.pbr.2016.04.017 2016

  6. [6]

    inary classification of hand mo ement directions from EEG using wavelet phase -locking,

    T. Chouhan, N. Robinson, A. P. Vinod, and K. K. Ang, “ inary classification of hand mo ement directions from EEG using wavelet phase -locking,” in 2017 IEEE International Conference on Systems, Man, and Cybernetics, SMC 2017 , Institute of Electrical and Electronics Engineers Inc., Nov. 2017, pp. 264 –269. doi: 10.1109/SMC.2017.8122613

    arXiv 2017

  7. [7]

    Wa let phase -locking based binary classification of hand mo ement directions from ,

    T. Chouhan, N. Robinson, A. P. Vinod, K. K. Ang, and . uan, “Wa let phase -locking based binary classification of hand mo ement directions from ,” J Neural Eng , vol. 15, no. 6, p. aadeed, 2018, doi: 10.1088/1741-2552/aadeed

    work page doi:10.1088/1741-2552/aadeed 2018

  8. [8]

    lassification of hand mo ement direction based on EEG high -gamma acti ity,

    C. A. Loza, G. R. Philips, M. K. Hazrati, J. J. Daly, and J. . rincipe, “ lassification of hand mo ement direction based on EEG high -gamma acti ity,” 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC 2014, pp. 6509 –6512, Nov. 2014, doi: 10.1109/EMBC.2014.6945119. Fig. 6. Filter rank assigned at A) La...

    work page arXiv 2014

Show all 32 references
  1. [9]

    Ma imum Variance-based EEG Time Bin Selection for Decoding of Imagined Hand Movement Directions in rain omputer Interface,

    K. Sagila Gangadharan, V. K. Benzy, and A. P. Vinod, “Ma imum Variance-based EEG Time Bin Selection for Decoding of Imagined Hand Movement Directions in rain omputer Interface,” BioSMART 2021 - Proceedings: 4th International Conference on Bio - Engineering for Smart Technologi...

    2021

  2. [10]

    Motor Imagery Hand Mo ement Direction Decoding using Brain Computer Interface to id Stroke eco ery and ehabilitation,

    V. Benzy, A. P. Vinod, R. Subasree, S. Alladi, and K. agha endra, “Motor Imagery Hand Mo ement Direction Decoding using Brain Computer Interface to id Stroke eco ery and ehabilitation,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 4320, no. c, pp. 1...

    2020

  3. [11]

    nline Hand Motor Imagery Direction Decoding using Brain Computer Interface,

    K. arashi a and . . Vinod, “ nline Hand Motor Imagery Direction Decoding using Brain Computer Interface,” Conf Proc IEEE Int Conf Syst Man Cybern , pp. 3182 –3187, 2021, doi: 10.1109/SMC52423.2021.9659302

    2021

  4. [12]

    ew hannel Selection Method using Autoencoder for Motor Imagery based rain omputer Interface,

    K. arashi a and . . Vinod, “ ew hannel Selection Method using Autoencoder for Motor Imagery based rain omputer Interface,” 2019 IEEE International Conference on Systems, Man and Cybernetics (SMC) , pp. 3641 –3646, Oct. 2019, doi: 10.1109/SMC.2019.8914251

    2019

  5. [13]

    re iew of channel selection algorithms for signal processing,

    T. Alotaiby, F. E. A. El -Samie, S. A. Alshebeili, and I. hmad, “ re iew of channel selection algorithms for signal processing,” EURASIP J Adv Signal Process, vol. 2015, no. 1, pp. 1 –21, Dec. 2015, doi: 10.1186/S13634-015-0251-9/TABLES/3

    2015 doi

  6. [14]

    Most opular Signal rocessing Methods in Motor - Imagery BCI: A Review and Meta - nalysis,

    Wier gała, D. Zapała, . M. Wojcik, and J. Masiak, “Most opular Signal rocessing Methods in Motor - Imagery BCI: A Review and Meta - nalysis,” Front Neuroinform, vol. 12, p. 78, Nov. 2018, doi: 10.3389/fninf.2018.00078

    2018

  7. [15]

    re iew of classification algorithms for EEG-based brain -computer interfaces: A 10 year update,

    F. Lotte et al., “ re iew of classification algorithms for EEG-based brain -computer interfaces: A 10 year update,” . doi . / -2552/aab2f2

  8. [16]

    Goodfellow, Y

    I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. MIT Press, 2016

    2016

  9. [17]

    Deep learning with convolutional neural networks for EEG decoding and isuali ation,

    R. T. Schirrmeister et al. , “Deep learning with convolutional neural networks for EEG decoding and isuali ation,” Hum Brain Mapp , vol. 38, no. 11, pp. 5391–5420, Nov. 2017, doi: 10.1002/HBM.23730

    2017 doi

  10. [18]

    et Compact Convolutional Neural Network for EEG -based Brain- omputer Interfaces,

    V. J. Lawhern, A. J. Solon, N. R. Waytowich, S. M. ordon, . . Hung, and . J. Lance, “ et Compact Convolutional Neural Network for EEG -based Brain- omputer Interfaces,” , ccessed ug. ,

  11. [19]

    F et: a Multi-view CNN with novel Variance Layer for Motor Imagery rain omputer Interface,

    R. Mane, N. Robinson, A. P. Vinod, S. -W. Lee, and C. uan, “F et: a Multi-view CNN with novel Variance Layer for Motor Imagery rain omputer Interface,” 42th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). IEEE, no. Mi, 2020

    2020

  12. [20]

    F MS et Filter -Bank Multi -Scale Convolutional Neural Network for EEG -Based Motor Imagery Decoding,

    K. Liu, M. Yang, Z. Yu, G. Wang, and W. Wu, “F MS et Filter -Bank Multi -Scale Convolutional Neural Network for EEG -Based Motor Imagery Decoding,” IEEE Trans Biomed Eng , vol. 70, no. 2, pp. 436–445, Feb. 2023, doi: 10.1109/TBME.2022.3193277

    2023

  13. [21]

    Deep learning -based electroencephalography analysis a systematic re iew,

    D. Steyrl et al. , “Deep learning -based electroencephalography analysis a systematic re iew,” J Neural Eng, vol. 16, no. 5, p. 051001, Aug. 2019, doi: 10.1088/1741-2552/AB260C

    2019 doi

  14. [22]

    Deep Transfer Learning for EEG - ased rain omputer Interface,

    Tan, F. Sun, and W. Zhang, “Deep Transfer Learning for EEG - ased rain omputer Interface,” ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings, vol. 2018 -April, pp. 916–920, Sep. 2018, doi: 10.1109/ICASSP.2018.8462115

    2018

  15. [23]

    Deep learning for motor imagery -based classification re iew,

    A. Al -Saegh, S. A. Dawwd, and J. M. Abdul -Jabbar, “Deep learning for motor imagery -based classification re iew,” Biomed Signal Process Control, vol. 63, p. 102172, Jan. 2021, doi: 10.1016/J.BSPC.2020.102172

    2021

  16. [24]

    Transfer Learning for - Based Brain-Computer Interfaces: A Review of Progress Made since ,

    D. Wu, Y. Xu, and . L. Lu, “Transfer Learning for - Based Brain-Computer Interfaces: A Review of Progress Made since ,” IEEE Trans Cogn Dev Syst , vol. 14, no. 1, pp. 4 –19, Mar. 2022, doi: 10.1109/TCDS.2020.3007453

    2022

  17. [25]

    Transfer Learning for rain - Computer Interfaces: A Euclidean Space Data Alignment pproach,

    H. He and D. Wu, “Transfer Learning for rain - Computer Interfaces: A Euclidean Space Data Alignment pproach,” IEEE Trans Biomed Eng , vol. 67, no. 2, pp. 399–410, Feb. 2020, doi: 10.1109/TBME.2019.2913914

    2020

  18. [26]

    Decoding Multi -Class Motor Imagery from Unilateral Limbs Using EEG Signals,

    F. ong, . Yang, and . uan, “Decoding Multi -Class Motor Imagery from Unilateral Limbs Using EEG Signals,” IEEE Transactions on Neural Systems and Rehabilitation Engineering , 2024, doi: 10.1109/TNSRE.2024.3454088

    2024

  19. [27]

    I ompetition – ra data set ,

    Leeb, . runner, … . M.-P.-G. U. of, and undefined , “ I ompetition – ra data set ,” lampx.tugraz.at, Accessed: Nov. 29, 2024. [Online]. Available: https://lampx.tugraz.at/~bci/database/004 - 2014/description.pdf

    2024

  20. [28]

    e iew of the I competition IV,

    M. Tangermann et al., “ e iew of the I competition IV,” Front Neurosci , vol. 6, no. JULY, p. 21084, Jul. 2012, doi: 10.3389/FNINS.2012.00055/BIBTEX

    2012 arXiv

  21. [29]

    Squee e -and-Excitation etworks,

    J. Hu, L. Shen, and . Sun, “Squee e -and-Excitation etworks,”

  22. [30]

    Impro ing direction decoding accuracy during online motor imagery based brain-computer interface using error -related potentials,

    K. arashi a and . . Vinod, “Impro ing direction decoding accuracy during online motor imagery based brain-computer interface using error -related potentials,” Biomed Signal Process Control, vol. 74, p. 103515, Apr. 2022, doi: 10.1016/J.BSPC.2022.103515

    2022

  23. [31]

    n analysis of performance evaluation for motor-imagery based I,

    Thomas, M. Dyson, and M. lerc, “ n analysis of performance evaluation for motor-imagery based I,” J Neural Eng , vol. 10, no. 3, p. 031001, Jun. 2013, doi: 10.1088/1741-2560/10/3/031001

    2013 doi

  24. [2024]

    Available: https://github.com/vlawhern/arl-eegmodels

    [Online]. Available: https://github.com/vlawhern/arl-eegmodels

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.