pith. sign in

arxiv: 2407.16249 · v1 · submitted 2024-07-23 · 🧬 q-bio.NC · eess.SP

How Does a Single EEG Channel Tell Us About Brain States in Brain-Computer Interfaces ?

Zaineb Ajra , Binbin Xu , G\'erard Dray , Jacky Montmain , St\'ephane Perrey This is my paper

Pith reviewed 2026-05-23 23:08 UTC · model grok-4.3

classification 🧬 q-bio.NC eess.SP
keywords EEGbrain-computer interfacesingle-channelconvolutional neural networkmental arithmeticmotor imagerybrain states
0
0 comments X

The pith

Single EEG channels classify mental tasks at up to 100% accuracy using cross-channel CNN strategies.

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

The paper tests whether brain-computer interfaces can move from many electrodes to a single channel for everyday use. It trains convolutional neural networks on data from multiple channels and tests on one channel at a time, then reverses the process by training on one channel and testing on the others. Lightweight networks that learn spectral-temporal features handle the classification of mental arithmetic and motor imagery. The approaches are checked on three public datasets and reach peak accuracies of 100 percent, 91.55 percent, and 73.45 percent in binary and three-class tasks on selected channels. A sympathetic reader would care because this could make portable, low-cost BCIs practical outside controlled labs.

Core claim

The paper claims that two distinct cross-channel strategies combined with few-parameter CNNs that extract fast spectral-temporal features enable single-channel EEG to classify brain states, demonstrated by highest accuracies of 100 percent, 91.55 percent and 73.45 percent in binary and 3-class settings on specific channels across three datasets of mental arithmetic and motor imagery tasks.

What carries the argument

Two cross-channel training/testing strategies paired with lightweight CNNs that learn spectral-temporal features from EEG.

If this is right

  • Single-channel devices become viable for classifying mental arithmetic and motor imagery outside laboratory settings.
  • Both multi-to-single and single-to-multi channel workflows support the same CNN architecture.
  • Portable BCI hardware can rely on one electrode while retaining usable performance on specific channels.
  • The methods provide a pathway to lower-cost EEG biomarkers for brain-state detection.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Hardware simplification could expand BCI use to consumer or home health settings where full caps are impractical.
  • The same cross-channel logic might transfer to other single-sensor modalities such as wearable heart-rate or eye-tracking data.
  • Long-term stability of the learned features across days or weeks remains an open test for real deployment.

Load-bearing premise

The reported peak accuracies on chosen channels will hold for truly unseen single-channel recordings without further subject-specific tuning or channel selection.

What would settle it

Classification accuracy falling below practical thresholds when the same models are applied to single-channel data from new subjects or recording sessions without retraining.

Figures

Figures reproduced from arXiv: 2407.16249 by Binbin Xu, G\'erard Dray, Jacky Montmain, St\'ephane Perrey, Zaineb Ajra.

Figure 1
Figure 1. Figure 1: The workflow of the MA-EEG signal classification process in this [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: CNN model architecture D. Training options and performance evaluation The proposed CNN model was trained with the following configuration: Stochastic gradient descent with momentum (SGDM) was used to optimize the model. The initial learning rate was fixed at 0.001. The maximum number of training epochs was 100, with a batch size of 64. In a previous study [15], the early stopping rule was strictly applied … view at source ↗
Figure 3
Figure 3. Figure 3: Topological representation of accuracy across channels for each dataset [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
read the original abstract

Over recent decades, neuroimaging tools, particularly electroencephalography (EEG), have revolutionized our understanding of the brain and its functions. EEG is extensively used in traditional brain-computer interface (BCI) systems due to its low cost, non-invasiveness, and high temporal resolution. This makes it invaluable for identifying different brain states relevant to both medical and non-medical applications. Although this practice is widely recognized, current methods are mainly confined to lab or clinical environments because they rely on data from multiple EEG electrodes covering the entire head. Nonetheless, a significant advancement for these applications would be their adaptation for "real-world" use, using portable devices with a single-channel. In this study, we tackle this challenge through two distinct strategies: the first approach involves training models with data from multiple channels and then testing new trials on data from a single channel individually. The second method focuses on training with data from a single channel and then testing the performances of the models on data from all the other channels individually. To efficiently classify cognitive tasks from EEG data, we propose Convolutional Neural Networks (CNNs) with only a few parameters and fast learnable spectral-temporal features. We demonstrated the feasibility of these approaches on EEG data recorded during mental arithmetic and motor imagery tasks from three datasets. We achieved the highest accuracies of 100%, 91.55% and 73.45% in binary and 3-class classification on specific channels across three datasets. This study can contribute to the development of single-channel BCI and provides a robust EEG biomarker for brain states classification.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes two cross-channel strategies for single-EEG-channel brain-state classification in BCIs using lightweight CNNs with spectral-temporal features: (1) multi-channel training followed by single-channel testing, and (2) single-channel training followed by testing on other channels. It evaluates these on mental arithmetic and motor imagery tasks from three datasets and reports peak accuracies of 100%, 91.55%, and 73.45% in binary and 3-class settings on specific channels, claiming feasibility for portable single-channel BCIs.

Significance. If the accuracies reflect performance on fixed, pre-specified single channels (rather than post-hoc maxima) and generalize without subject-specific tuning, the work would support more accessible, low-electrode BCIs. The empirical nature of the study provides no parameter-free derivations or machine-checked proofs, but reproducible code or per-channel performance tables would strengthen its contribution to single-channel BCI development.

major comments (3)
  1. [Abstract] Abstract: The reported 'highest accuracies of 100%, 91.55% and 73.45% ... on specific channels' are presented as maxima across channels after evaluation. This post-hoc selection step is not part of either proposed cross-channel strategy and would be unavailable for true single-channel deployment on unseen data; the headline numbers therefore do not demonstrate reliable performance on any fixed channel as required by the central claim.
  2. [Methods and Results] Methods/Results: No information is supplied on the number of subjects per dataset, the cross-validation procedure (e.g., subject-wise or trial-wise), statistical testing, error bars, or whether the 'specific channels' were pre-specified before testing. These omissions make it impossible to assess whether the reported peaks support generalization to unseen single-channel data.
  3. [Results] Results: The two strategies are intended to test generalization to unseen single channels, yet the abstract and results emphasize peak values on selected channels. If the per-channel accuracies (without selection) fall substantially below the reported maxima, the feasibility claim for single-channel BCIs would require substantial revision.
minor comments (2)
  1. [Abstract] The abstract and introduction would benefit from explicit statements of the number of subjects and datasets used, rather than the generic phrase 'three datasets'.
  2. [Methods] Notation for the CNN architecture (number of layers, filter sizes, input dimensions) should be defined consistently between text and any accompanying figure or table.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback, which helps clarify the presentation of our cross-channel strategies. We address each major comment below and will revise the manuscript to improve transparency and support for the single-channel BCI feasibility claim.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The reported 'highest accuracies of 100%, 91.55% and 73.45% ... on specific channels' are presented as maxima across channels after evaluation. This post-hoc selection step is not part of either proposed cross-channel strategy and would be unavailable for true single-channel deployment on unseen data; the headline numbers therefore do not demonstrate reliable performance on any fixed channel as required by the central claim.

    Authors: We agree that the reported peak values are maxima obtained after evaluating multiple channels. The two strategies themselves (multi-channel training with single-channel testing, and single-channel training with cross-channel testing) are designed to assess performance on individual channels without requiring all channels at inference time. To better align the abstract with this, we will revise it to report both the peak and the mean/median accuracies across channels under each strategy, and explicitly note that channel selection occurs during evaluation rather than being available for completely unseen fixed-channel deployment. This will temper the headline claims while preserving the demonstration of feasibility on promising channels. revision: yes

  2. Referee: [Methods and Results] Methods/Results: No information is supplied on the number of subjects per dataset, the cross-validation procedure (e.g., subject-wise or trial-wise), statistical testing, error bars, or whether the 'specific channels' were pre-specified before testing. These omissions make it impossible to assess whether the reported peaks support generalization to unseen single-channel data.

    Authors: These details were inadvertently omitted from the initial submission. We will add a dedicated Experimental Setup subsection stating the subject counts from each public dataset, the cross-validation scheme (subject-independent where appropriate to test generalization), the use of statistical tests (e.g., paired t-tests across folds), error bars on all accuracy plots, and confirmation that channels were not pre-specified but evaluated exhaustively. This will allow readers to judge the strength of generalization evidence directly. revision: yes

  3. Referee: [Results] Results: The two strategies are intended to test generalization to unseen single channels, yet the abstract and results emphasize peak values on selected channels. If the per-channel accuracies (without selection) fall substantially below the reported maxima, the feasibility claim for single-channel BCIs would require substantial revision.

    Authors: The strategies explicitly evaluate generalization to individual channels; the peaks simply illustrate the best attainable performance when a suitable channel is available. We will expand the Results section with complete per-channel accuracy tables (or heatmaps) for both strategies across all datasets. This will enable direct assessment of typical versus peak performance. If the non-selected accuracies are substantially lower, we will add discussion on the practical need for channel pre-selection or calibration in portable BCI applications. revision: partial

Circularity Check

0 steps flagged

Empirical classification study with no derivation chain

full rationale

This paper reports experimental results from training and evaluating CNN models on three EEG datasets for mental arithmetic and motor imagery classification tasks. The two cross-channel strategies are described as training/testing procedures, and the accuracies (including reported peaks) are direct outputs of those experiments on the data. No mathematical derivation, first-principles result, or prediction is claimed that reduces to its own inputs by construction. There are no load-bearing self-citations, ansatzes, or uniqueness theorems invoked. The work is self-contained as a data-driven empirical demonstration and does not exhibit any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; CNN training implicitly involves many hyperparameters but none are named or fitted in the provided text.

pith-pipeline@v0.9.0 · 5833 in / 1159 out tokens · 19368 ms · 2026-05-23T23:08:54.641456+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

21 extracted references · 21 canonical work pages

  1. [1]

    Brain-computer interfaces (bcis) for communication and control,

    J. R. Wolpaw, “Brain-computer interfaces (bcis) for communication and control,” in Proceedings of the 9th international ACM SIGACCESS conference on Computers and accessibility , pp. 1–2, 2007

    work page 2007

  2. [2]

    Bci in patients with disorders of consciousness: clinical perspectives,

    J. Luaut ´e, D. Morlet, and J. Mattout, “Bci in patients with disorders of consciousness: clinical perspectives,” Annals of Physical and Rehabili- tation Medicine, vol. 58, no. 1, pp. 29–34, 2015

    work page 2015

  3. [3]

    Neurofeedback treatment for attention-deficit/hyperactivity disorder in children: a comparison with methylphenidate,

    T. Fuchs, N. Birbaumer, W. Lutzenberger, J. H. Gruzelier, and J. Kaiser, “Neurofeedback treatment for attention-deficit/hyperactivity disorder in children: a comparison with methylphenidate,” Applied psychophysiol- ogy and biofeedback , vol. 28, no. 1, pp. 1–12, 2003

    work page 2003

  4. [4]

    A review of classification algorithms for eeg-based brain–computer interfaces,

    F. Lotte, M. Congedo, A. L ´ecuyer, F. Lamarche, and B. Arnaldi, “A review of classification algorithms for eeg-based brain–computer interfaces,” Journal of neural engineering , vol. 4, no. 2, p. R1, 2007

    work page 2007

  5. [5]

    A biologically inspired approach to frequency domain feature extraction for eeg classification,

    N. G. Ozmen, L. Gumusel, and Y . Yang, “A biologically inspired approach to frequency domain feature extraction for eeg classification,” Computational and mathematical methods in medicine , vol. 2018, 2018

    work page 2018

  6. [6]

    A comparative study of features and classifiers in single-channel eeg-based motor imagery bci,

    S. Kanoga, A. Kanemura, and H. Asoh, “A comparative study of features and classifiers in single-channel eeg-based motor imagery bci,” in 2018 IEEE Global Conference on Signal and Information Processing (GlobalSIP), pp. 474–478, IEEE, 2018

    work page 2018

  7. [7]

    Assessing focus through ear-eeg: a comparative study between conventional cap eeg and mobile in-and around-the-ear eeg systems,

    G. Cr ´etot-Richert, M. De V os, S. Debener, M. G. Bleichner, and J. V oix, “Assessing focus through ear-eeg: a comparative study between conventional cap eeg and mobile in-and around-the-ear eeg systems,” Frontiers in Neuroscience, vol. 17, p. 895094, 2023

    work page 2023

  8. [8]

    A full upper limb robotic exoskeleton for reaching and grasping rehabilitation triggered by mi-bci,

    M. Barsotti, D. Leonardis, C. Loconsole, M. Solazzi, E. Sotgiu, C. Pro- copio, C. Chisari, M. Bergamasco, and A. Frisoli, “A full upper limb robotic exoskeleton for reaching and grasping rehabilitation triggered by mi-bci,” in 2015 IEEE international conference on rehabilitation robotics (ICORR), pp. 49–54, IEEE, 2015

    work page 2015

  9. [9]

    Patients’ eeg data analysis via spectrogram image with a convolution neural network,

    L. Yuan and J. Cao, “Patients’ eeg data analysis via spectrogram image with a convolution neural network,” in Intelligent Decision Technolo- gies 2017: Proceedings of the 9th KES International Conference on Intelligent Decision Technologies (KES-IDT 2017)–Part I 9 , pp. 13–21, Springer, 2018

    work page 2017

  10. [10]

    Deep learning vs. traditional computer vision,

    N. O’Mahony, S. Campbell, A. Carvalho, S. Harapanahalli, G. V . Hernandez, L. Krpalkova, D. Riordan, and J. Walsh, “Deep learning vs. traditional computer vision,” in Advances in Computer Vision: Proceedings of the 2019 Computer Vision Conference (CVC), Volume 1 1, pp. 128–144, Springer, 2020

    work page 2019

  11. [11]

    Single-channel eeg classification by multi-channel tensor subspace learning and regression,

    S. Van Eyndhoven, M. Bouss ´e, B. Hunyadi, L. De Lathauwer, and S. Van Huffel, “Single-channel eeg classification by multi-channel tensor subspace learning and regression,” in 2018 IEEE 28th International Workshop on Machine Learning for Signal Processing (MLSP), pp. 1–6, IEEE, 2018

    work page 2018

  12. [12]

    Development of single-channel hybrid bci system using motor imagery and ssvep,

    L.-W. Ko, S. Ranga, O. Komarov, and C.-C. Chen, “Development of single-channel hybrid bci system using motor imagery and ssvep,” Journal of healthcare engineering , vol. 2017, 2017

    work page 2017

  13. [13]

    Classification of four-class motor imagery employing single-channel electroencephalography,

    S. Ge, R. Wang, and D. Yu, “Classification of four-class motor imagery employing single-channel electroencephalography,” PloS one , vol. 9, no. 6, p. e98019, 2014

    work page 2014

  14. [14]

    Eyes-closed hybrid brain- computer interface employing frontal brain activation,

    J. Shin, K.-R. M ¨uller, and H.-J. Hwang, “Eyes-closed hybrid brain- computer interface employing frontal brain activation,”PloS one, vol. 13, no. 5, p. e0196359, 2018

    work page 2018

  15. [15]

    Mental arith- metic task classification with convolutional neural network based on spectral-temporal features from eeg,

    Z. Ajra, B. Xu, G. Dray, J. Montmain, and S. Perrey, “Mental arith- metic task classification with convolutional neural network based on spectral-temporal features from eeg,” in 2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), pp. 52–55, IEEE, 2022

    work page 2022

  16. [16]

    Open access dataset for eeg+nirs single-trial classification,

    J. Shin, A. von L ¨uhmann, B. Blankertz, D.-W. Kim, J. Jeong, H.-J. Hwang, and K.-R. M ¨uller, “Open access dataset for eeg+nirs single-trial classification,”IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 25, pp. 1735–1745, 10 2017

    work page 2017

  17. [17]

    Multi-channel eeg recording during motor imagery of different joints from the same limb,

    X. Ma, S. Qiu, and H. He, “Multi-channel eeg recording during motor imagery of different joints from the same limb,” Scientific data, vol. 7, no. 1, p. 191, 2020

    work page 2020

  18. [18]

    Automatic sleep stage scoring using time-frequency analysis and stacked sparse autoencoders,

    O. Tsinalis, P. M. Matthews, and Y . Guo, “Automatic sleep stage scoring using time-frequency analysis and stacked sparse autoencoders,” Annals of biomedical engineering , vol. 44, pp. 1587–1597, 2016

    work page 2016

  19. [19]

    Stacked auto-encoder based tagging with deep features for content-based medical image retrieval,

    S ¸. ¨Ozt¨urk, “Stacked auto-encoder based tagging with deep features for content-based medical image retrieval,” Expert Systems with Applica- tions, vol. 161, p. 113693, 2020. 6

    work page 2020

  20. [20]

    A novel classification framework us- ing multiple bandwidth method with optimized cnn for brain–computer interfaces with eeg-fnirs signals,

    M. Nour, S ¸. ¨Ozt¨urk, and K. Polat, “A novel classification framework us- ing multiple bandwidth method with optimized cnn for brain–computer interfaces with eeg-fnirs signals,” Neural Computing and Applications , vol. 33, no. 22, pp. 15815–15829, 2021

    work page 2021

  21. [21]

    Feature extraction and classification methods for hybrid fnirs-eeg brain-computer interfaces,

    K.-S. Hong, M. J. Khan, and M. J. Hong, “Feature extraction and classification methods for hybrid fnirs-eeg brain-computer interfaces,” Frontiers in Human Neuroscience , vol. 12, p. 246, 2018. 7

    work page 2018