pith. machine review for the scientific record. sign in

arxiv: 2605.06018 · v1 · submitted 2026-05-07 · 💻 cs.HC

Recognition: unknown

I see artifacts: ICA-based EEG artifact removal does not improve deep network decoding across three BCI tasks

Taeho Kang , Yiyu Chen , Christian Wallraven
Authors on Pith no claims yet

Pith reviewed 2026-05-08 07:26 UTC · model grok-4.3

classification 💻 cs.HC
keywords EEGICA artifact removaldeep learning decodingBCI tasksneural network performancecomponent rejection
0
0 comments X

The pith

ICA-based artifact removal in EEG signals does not consistently improve deep network decoding across BCI tasks

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

The paper tests whether independent component analysis (ICA) for artifact rejection in EEG data helps deep neural networks decode brain signals more accurately in brain-computer interface applications. It runs a full matrix of two ICA methods, three rejection strategies including no rejection, and three common network architectures on three separate datasets covering motor imagery, memory formation, and visual memory tasks. Performance is measured through within-participant and within-dataset cross-validation. The results indicate that rejected data yields at best only minor gains over unrejected data, despite the added computational cost of ICA.

Core claim

Component-rejected EEG data produced by Infomax or AMICA combined with ICLabel or MARA does not deliver consistently higher decoding accuracy in CNN or LSTM models than the same data without any rejection, across motor imagery, long-term memory, and visual memory tasks.

What carries the argument

A pipeline matrix that applies ICA decomposition followed by optional component rejection before feeding the signals into neural network classifiers, then compares accuracy with and without the rejection step.

If this is right

  • Deep networks can achieve comparable decoding performance on raw EEG without explicit ICA-based cleaning.
  • The computational overhead of running ICA may not be justified when the goal is only to improve classification accuracy.
  • Within-participant cross-validation results generalize similarly whether or not artifact rejection is applied.

Where Pith is reading between the lines

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

  • In practice, BCI developers could skip ICA preprocessing to reduce latency in real-time systems.
  • Attention may shift toward other preprocessing choices or architecture tweaks that more reliably affect performance.
  • The finding raises the question of whether ICA still adds value when combined with newer model types or larger datasets.

Load-bearing premise

The three chosen datasets, two ICA algorithms, three rejection strategies, and three network architectures are representative of typical BCI decoding scenarios.

What would settle it

A follow-up experiment on new EEG datasets that finds statistically significant and consistent accuracy gains from the same ICA rejection pipelines would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.06018 by Christian Wallraven, Taeho Kang, Yiyu Chen.

Figure 1
Figure 1. Figure 1: Study overview dynamics as well as to investigate the possibility of a generalized result, the BCI competition dataset IV 2a (Motor Imagery ) [88], a long-term memory prediction dataset [63], and a visual memory dataset from EEGManyPipelines [64, 89] were used. The BCI competition dataset IV 2a consists of 9 participants worth of data recorded with 22-EEG channels (+3 EOG), with one training session and on… view at source ↗
Figure 2
Figure 2. Figure 2: Visualization of post-hoc Tukey’s HSD test performed on each participant’s entire gridsearch space validation F1 for a) Shallownet, b) MLSTM-FCN, c) EEGNet architectures on the BCIC-IV2a dataset. Each row denotes individual participant in the dataset, while each column represents a single comparison pair among the 5 ICA component rejection pipelines. Cells are shaded in gray if the pair comparison did not … view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of post-hoc Tukey’s HSD test performed on each participant’s entire gridsearch space validation F1 for a) Shallownet, b) MLSTM-FCN, c) EEGNet architectures on the long-term memory dataset. Each row denotes individual participant in the dataset, while each column represents a single comparison pair among the 5 ICA component rejection pipelines. Cells are shaded in gray if the pair comparison d… view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of post-hoc Tukey’s HSD test performed on each participant’s entire gridsearch space validation F1 for a) Shallownet, b) MLSTM-FCN, c) EEGNet architectures on the visual memory dataset. Each row denotes individual participant in the dataset, while each column represents a single comparison pair among the 5 ICA component rejection pipelines. Cells are shaded in gray if the pair comparison did … view at source ↗
Figure 5
Figure 5. Figure 5: Across network and pipelines correlational matrix of the best-parameter validation F1 scores (green) and holdout test-set F1 scores (red) for the BCIC-IV2a dataset. Pipeline labels are abbreviated for space. NoICA R_MARA R_ICLabel A_MARA A_ICLabel NoICA R_MARA R_ICLabel A_MARA A_ICLabel NoICA R_MARA R_ICLabel A_MARA A_ICLabel NoICA R_MARA R_ICLabel A_MARA A_ICLabel NoICA R_MARA R_ICLabel A_MARA A_ICLabel N… view at source ↗
Figure 6
Figure 6. Figure 6: Across network and pipelines correlational matrix of the best-parameter validation F1 scores (green) and holdout test-set F1 scores (red) for the long-term memory dataset. Pipeline labels are abbreviated for space view at source ↗
Figure 7
Figure 7. Figure 7: Across network and pipelines correlational matrix of the best-parameter validation F1 scores (green) and holdout test-set F1 scores (red) for the visual memory dataset. Pipeline labels are abbreviated for space. p=.029) and long-term memory dataset (Shal￾lowNet F(dfgrp4, dfrm56)=5.432, p=.001; MLSTM￾FCN F=1.151, p=.338, EEGNetV4 F=3.203, p=.019). In a post-hoc analysis with paired T-tests corrected by Bonf… view at source ↗
Figure 8
Figure 8. Figure 8: Overall test-set F1 score boxplots for each pipeline for each network for a) BCIC-IV2a, b) Long-term memory, c) Visual memory datasets. Significance markings of pipeline comparisons are based on post-hoc paired T-tests corrected with Bonferroni correction where the corrected p-value was lower than 0.05. Whiskers are defined as higher and lower quantiles of the data, and the center line denotes the median. … view at source ↗
Figure 9
Figure 9. Figure 9: Scatterplots of number of components rejected and the test-F1 scores in each dataset-model combination given for each participant in dataset. X-axis shows the number of components rejected as a result of ICA-based noise rejections, Y-axis shows the test-F1 scores of each participant. The X-values are slightly jittered for visual readability as noICA pipelines all have 0 for component rejection values. Row … view at source ↗
Figure 10
Figure 10. Figure 10: Participant F1 scores for all pipelines, with participant number re-sorted by NoICA pipeline F1 scores. seen in Figures 5, 6, 7. Are these findings transferable to standard machine learning models? view at source ↗
Figure 11
Figure 11. Figure 11: LDA results of the BCICIV2a dataset, performed on raw EEG waveforms without additional feature extraction that are standard for motor imagery. convolutional layer of the best performing EEGNetV4‡ models from each pipeline using the test-set data of the first participant in the BCIC-IV2a dataset. The averaged CAM heatmap of the test-set for each pipeline can be found in view at source ↗
Figure 12
Figure 12. Figure 12: Averaged GradCAM values from the BCICIV2a dataset, participant 1, test set from the EEGNetV4’s best parameter model of each pipeline. Values from the conv-separable-depth layer were computed. a) SSIM-EEG NoICA RunICA-MARA RunICA-ICLabel AMICA-MARA AMICA-ICLabel NoICA 1 0.8031 0.9374 0.7989 1 RunICA-MARA 0.8031 1 0.7772 0.9695 0.8031 RunICA-ICLabel 0.9374 0.7772 1 0.7716 0.9374 AMICA-MARA 0.7989 0.9695 0.7… view at source ↗
Figure 13
Figure 13. Figure 13: Comparison of Raw EEG waveforms of a test set trial from the BCICIV2a dataset, participant 1, for each pipeline. GradCAM values of the best performing EEGNetV4 model are shown as color gradients on the waveforms. tested BCIC-IV2a dataset. Considering that many recent SOTA studies in motor imagery classification have used task-specific feature extraction methods [111–114] (see [115] for review), this is no… view at source ↗
read the original abstract

In this paper, we conduct a detailed investigation on the effect of independent component (IC)-based noise rejection methods in neural network classifier-based decoding of electroencephalography (EEG) data in different task datasets. We apply a pipeline matrix of two popular different independent component (IC) decomposition methods (Infomax and Adaptive Mixture Independent Component Analysis (AMICA)) with three different component rejection strategies (none, ICLabel, and multiple artifact rejection algorithm [MARA]) on three different EEG datasets (motor imagery, long-term memory formation, and visual memory). We cross-validate processed data from each pipeline with three architectures commonly used for EEG classification (two convolutional neural networks and one long short-term memory-based model. We compare decoding performances on within-participant and within-dataset levels.Our results show that the benefit from using IC-based noise rejection for decoding analyses is at best minor, as component-rejected data did not show consistently better performance than data without rejections; especially given the significant computational resources required for independent component analysis (ICA) computations.

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

1 major / 0 minor

Summary. The paper conducts a systematic empirical evaluation of ICA-based artifact removal for deep neural network-based EEG decoding. It tests two ICA decomposition methods (Infomax and AMICA), three rejection strategies (none, ICLabel, MARA), across three datasets (motor imagery, long-term memory formation, visual memory), and three network architectures (two CNNs and one LSTM). Using within-participant and within-dataset cross-validation, the central finding is that data with component rejection does not consistently outperform data without rejection, indicating that the benefit of IC-based noise rejection is at best minor, particularly in light of the computational demands of ICA.

Significance. This negative result, if substantiated, is significant for the BCI and EEG machine learning community as it challenges the routine application of ICA preprocessing in deep learning pipelines for EEG classification. The strength lies in the broad experimental matrix covering multiple methods, datasets, and models, providing a more robust test than single-dataset studies. The authors deserve credit for the comprehensive design, the use of held-out cross-validation, and the focus on practical implications regarding computational resources. This could encourage reevaluation of preprocessing steps in resource-constrained BCI applications.

major comments (1)
  1. Results section: The claim that component-rejected data 'did not show consistently better performance' and that the benefit is 'at best minor' is central but unsupported by reported effect sizes, mean accuracy deltas, standard deviations, or statistical tests (e.g., paired t-tests or Wilcoxon tests across pipelines). Without these in tables or text for each dataset-architecture combination, the strength of the negative conclusion cannot be fully assessed.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive review and positive assessment of the manuscript's scope and design. We agree that the central negative finding requires stronger quantitative backing and will revise the Results section to include the requested metrics and tests.

read point-by-point responses

  1. Referee: Results section: The claim that component-rejected data 'did not show consistently better performance' and that the benefit is 'at best minor' is central but unsupported by reported effect sizes, mean accuracy deltas, standard deviations, or statistical tests (e.g., paired t-tests or Wilcoxon tests across pipelines). Without these in tables or text for each dataset-architecture combination, the strength of the negative conclusion cannot be fully assessed.

    Authors: We accept this criticism. In the revised manuscript we will add a table (main text or supplementary) that reports, for every dataset–architecture pair, the mean accuracy and standard deviation under each preprocessing pipeline (none, ICLabel, MARA) for both Infomax and AMICA decompositions. We will also compute and report mean accuracy deltas relative to the no-rejection baseline and will apply paired non-parametric tests (Wilcoxon signed-rank) across participants or folds, together with effect-size measures (e.g., rank-biserial correlation). These additions will allow readers to evaluate both the magnitude and statistical reliability of any observed differences. revision: yes

Circularity Check

0 steps flagged

No significant circularity: purely empirical comparison

full rationale

The paper reports an empirical matrix of ICA pipelines (Infomax/AMICA × none/ICLabel/MARA) applied to three EEG datasets, decoded by three neural architectures, with performance measured via within-participant and within-dataset cross-validation. No derivations, equations, fitted parameters renamed as predictions, or self-citation chains appear in the abstract or described methods. The central negative result (component-rejected data not consistently superior) follows directly from comparing measured accuracies on held-out folds; it does not reduce to any input by construction. This is a standard, self-contained empirical study with no load-bearing self-referential steps.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The paper is purely empirical. No free parameters are introduced in the central claim. Standard statistical assumptions of ICA (statistical independence of sources) and cross-validation are used but not invented by the authors.

axioms (2)
  • domain assumption Independent components recovered by Infomax and AMICA correspond to physiologically meaningful sources that can be labeled as artifacts.
    Invoked when applying ICLabel and MARA rejection strategies.
  • domain assumption Within-participant cross-validation on the chosen datasets provides a fair estimate of decoding performance.
    Used for all performance comparisons.

pith-pipeline@v0.9.0 · 5489 in / 1264 out tokens · 51312 ms · 2026-05-08T07:26:51.892437+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

124 extracted references · 5 canonical work pages · 2 internal anchors

  1. [1]

    Results 3.1. Component Rejections In the BCI Competition IV-2A dataset, an average of 0.1 artifact components were rejected from AMICA component data rejected with ICLabel, while an average of 4.9 artifacts components were rejected through MARA. In data processed with RunICA, an average of 0.1 components were rejected through ICLabel, and 5.1 through MARA...

  2. [2]

    10), while all 15 participants had significant p-values in MLSTM-FCN

    and Shallownet (p > .05 in ptc. 10), while all 15 participants had significant p-values in MLSTM-FCN. As can be seen in the circos plots, the chosen best performing parameters were highly variable between participants (and somewhat even across pipelines in same participant) for MLSTM-FCN and ShallowNet, while they were somewhat consistent for EEGNetV4 in ...

  3. [3]

    To see whether the results would be generalizable, we chose 3 different network architectures, including two specifically designed for brain signal classification

    Discussion In this study, we pre-processed 3 datasets of epoched EEG signals purposed for BCI classification through a total of 4 (+1 control) different ICA-based automatic pipelines for rejecting artifact components, to investigate whether such pre-processing procedures affect general classifier performance in neural network- based architectures. To see ...

  4. [4]

    2017-0-00451, Development of BCI based Brain and Cognitive Com- puting Technology for Recognizing User’s Intentions using Deep Learning; No

    Acknowledgments This study was supported by the National Re- search Foundation of Korea under project BK21 FOUR and grants NRF-2022R1A2C2092118, NRF- 2022R1H1A2092007, NRF-2019R1A2C2007612, as well as by Institute of Information & Communica- tions Technology Planning & Evaluation (IITP) grants funded by the Korea government (No. 2017-0-00451, Development ...

    work page doi:10.55776/i6222 2017

  5. [5]

    Committee report: publication guidelines and recommendations for studies using electroencephalography and magnetoencephalography,

    A. Keil, S. Debener, G. Gratton, M. Jungh¨ ofer, E. S. Kappenman, S. J. Luck, P. Luu, G. A. Miller, and C. M. Yee, “Committee report: publication guidelines and recommendations for studies using electroencephalography and magnetoencephalography,” Psychophysiology, vol. 51, no. 1, pp. 1–21, 2014

    2014

  6. [6]

    Brain-computer interface technologies: from signal to action,

    A. Ortiz-Rosario and H. Adeli, “Brain-computer interface technologies: from signal to action,”Reviews in the Neurosciences, vol. 24, no. 5, pp. 537–552, 2013

    2013

  7. [7]

    Good practice for conducting and reporting meg research,

    J. Gross, S. Baillet, G. R. Barnes, R. N. Henson, A. Hillebrand, O. Jensen, K. Jerbi, V. Litvak, B. Maess, R. Oostenveldet al., “Good practice for conducting and reporting meg research,”Neuroimage, vol. 65, pp. 349– 363, 2013

    2013

  8. [8]

    Preprocessing of eeg,

    S.-P. Kim, “Preprocessing of eeg,” inComputational EEG Analysis. Springer, 2018, pp. 15–33

    2018

  9. [9]

    The PREP pipeline: Standardized preprocessing for large-scale EEG analysis,

    N. Bigdely-Shamlo, T. Mullen, C. Kothe, K.-M. Su, and K. A. Robbins, “The PREP pipeline: Standardized preprocessing for large-scale EEG analysis,”Frontiers in Neuroinformatics, vol. 9, p. 16, 2015

    2015

  10. [10]

    Automated eeg mega- analysis i: Spectral and amplitude characteristics across studies,

    N. Bigdely-Shamlo, J. Touryan, A. Ojeda, C. Kothe, T. Mullen, and K. Robbins, “Automated eeg mega- analysis i: Spectral and amplitude characteristics across studies,”NeuroImage, vol. 207, p. 116361, 2020

    2020

  11. [11]

    Comparison of eeg preprocessing methods to improve the classification of p300 trials,

    Z. Cashero, “Comparison of eeg preprocessing methods to improve the classification of p300 trials,”Master of Science Thesis, Colorado State University, 2011

    2011

  12. [12]

    Compar- ison of different wavelet features from eeg signals for classifying human emotions,

    M. Murugappan, R. Nagarajan, and S. Yaacob, “Compar- ison of different wavelet features from eeg signals for classifying human emotions,” in2009 IEEE symposium on industrial electronics & applications, vol. 2. IEEE, 2009, pp. 836–841

    2009

  13. [13]

    To explore the potentials of independent component analysis in brain- computer interface of motor imagery,

    X. Wu, B. Zhou, Z. Lv, and C. Zhang, “To explore the potentials of independent component analysis in brain- computer interface of motor imagery,”IEEE Journal of Biomedical and Health Informatics, vol. 24, no. 3, pp. 775–787, 2019

    2019

  14. [14]

    The correction of eye blink artefacts in the eeg: a comparison of two prominent methods,

    S. Hoffmann and M. Falkenstein, “The correction of eye blink artefacts in the eeg: a comparison of two prominent methods,”PloS one, vol. 3, no. 8, p. e3004, 2008

    2008

  15. [15]

    Removing electroencephalographic artifacts: comparison between ica and pca,

    T.-P. Jung, C. Humphries, T.-W. Lee, S. Makeig, M. J. McKeown, V. Iragui, and T. J. Sejnowski, “Removing electroencephalographic artifacts: comparison between ica and pca,” inNeural Networks for Signal Processing VIII. Proceedings of the 1998 IEEE Signal Processing Society Workshop (Cat. No. 98TH8378). IEEE, 1998, pp. 63–72

    1998

  16. [16]

    The Harvard Automated Pro- cessing Pipeline for Electroencephalography (HAPPE): Standardized processing software for developmental and high-artifact data,

    L. J. Gabard-Durnam, A. S. Mendez Leal, C. L. Wilkin- son, and A. R. Levin, “The Harvard Automated Pro- cessing Pipeline for Electroencephalography (HAPPE): Standardized processing software for developmental and high-artifact data,”Frontiers in Neuroscience, vol. 12, p. 97, 2018

    2018

  17. [17]

    An automatic pre-processing pipeline for EEG Analysis (APP) based on robust statistics,

    J. R. da Cruz, V. Chicherov, M. H. Herzog, and P. Figueiredo, “An automatic pre-processing pipeline for EEG Analysis (APP) based on robust statistics,” Clinical Neurophysiology, vol. 129, no. 7, pp. 1427–1437, 2018

    2018

  18. [18]

    Automagic: Standardized preprocessing of big EEG data,

    A. Pedroni, A. Bahreini, and N. Langer, “Automagic: Standardized preprocessing of big EEG data,”Neu- roImage, 2019

    2019

  19. [19]

    Robust artifactual independent component classification for BCI practitioners,

    I. Winkler, S. Brandl, F. Horn, E. Waldburger, C. Allefeld, and M. Tangermann, “Robust artifactual independent component classification for BCI practitioners,”Jour- nal of Neural Engineering, vol. 11, no. 3, p. 035013, 2014

    2014

  20. [20]

    Iclabel: An automated electroencephalographic inde- pendent component classifier, dataset, and website,

    L. Pion-Tonachini, K. Kreutz-Delgado, and S. Makeig, “Iclabel: An automated electroencephalographic inde- pendent component classifier, dataset, and website,” NeuroImage, vol. 198, pp. 181–197, 2019

    2019

  21. [21]

    Investigation into the origin of the noise of surface electrodes,

    E. Huigen, A. Peper, and C. Grimbergen, “Investigation into the origin of the noise of surface electrodes,” Medical and biological engineering and computing, vol. 40, pp. 332–338, 2002

    2002

  22. [22]

    Adaptive digital notch filter design on the unit circle for the removal of powerline noise from biomedical signals,

    M. Ferdjallah and R. E. Barr, “Adaptive digital notch filter design on the unit circle for the removal of powerline noise from biomedical signals,”IEEE transactions on biomedical engineering, vol. 41, no. 6, pp. 529–536, 1994

    1994

  23. [23]

    Filters: when, why, and how (not) to use them,

    A. de Cheveign´ e and I. Nelken, “Filters: when, why, and how (not) to use them,”Neuron, vol. 102, no. 2, pp. 280–293, 2019

    2019

  24. [24]

    Removal of ocular artifacts from eeg using adaptive thresholding of wavelet coefficients,

    V. Krishnaveni, S. Jayaraman, L. Anitha, and K. Ra- madoss, “Removal of ocular artifacts from eeg using adaptive thresholding of wavelet coefficients,”Journal of neural engineering, vol. 3, no. 4, p. 338, 2006

    2006

  25. [25]

    Faster: fully automated statistical thresholding for eeg artifact rejection,

    H. Nolan, R. Whelan, and R. B. Reilly, “Faster: fully automated statistical thresholding for eeg artifact rejection,”Journal of neuroscience methods, vol. 192, I See Artifacts: ICA-based EEG Artifact Removal Does Not Improve Deep Network Decoding Across Three BCI Tasks22 no. 1, pp. 152–162, 2010

    2010

  26. [26]

    A generic eeg artifact removal algorithm based on the multi-channel wiener filter,

    B. Somers, T. Francart, and A. Bertrand, “A generic eeg artifact removal algorithm based on the multi-channel wiener filter,”Journal of neural engineering, vol. 15, no. 3, p. 036007, 2018

    2018

  27. [27]

    Optimal filters for erp research i: A general approach for selecting filter settings,

    G. Zhang, D. R. Garrett, and S. J. Luck, “Optimal filters for erp research i: A general approach for selecting filter settings,”Psychophysiology, p. e14531, 2024

    2024

  28. [28]

    Eeg artifact removal—state-of-the-art and guidelines,

    J. A. Urig¨ uen and B. Garcia-Zapirain, “Eeg artifact removal—state-of-the-art and guidelines,”Journal of neural engineering, vol. 12, no. 3, p. 031001, 2015

    2015

  29. [29]

    Univariate versus multivariate tests in repeated-measures experiments

    M. L. Davidson, “Univariate versus multivariate tests in repeated-measures experiments.”Psychological Bul- letin, vol. 77, no. 6, p. 446, 1972

    1972

  30. [30]

    Deconstructing multivariate decoding for the study of brain function,

    M. N. Hebart and C. I. Baker, “Deconstructing multivariate decoding for the study of brain function,” Neuroimage, vol. 180, pp. 4–18, 2018

    2018

  31. [31]

    The same analysis approach: Practical protection against the pitfalls of novel neuroimaging analysis methods,

    K. G¨ orgen, M. N. Hebart, C. Allefeld, and J.-D. Haynes, “The same analysis approach: Practical protection against the pitfalls of novel neuroimaging analysis methods,”Neuroimage, vol. 180, pp. 19–30, 2018

    2018

  32. [32]

    Does data cleaning improve brain state classification?

    S. L. Meisler, M. J. Kahana, and Y. Ezzyat, “Does data cleaning improve brain state classification?”Journal of neuroscience methods, vol. 328, p. 108421, 2019

    2019

  33. [33]

    Ica: a potential tool for bci systems,

    A. Kachenoura, L. Albera, L. Senhadji, and P. Comon, “Ica: a potential tool for bci systems,”IEEE Signal Processing Magazine, vol. 25, no. 1, pp. 57–68, 2007

    2007

  34. [34]

    Comparing results of algorithms implementing blind source separation of eeg data,

    A. Delorme, J. Plamer, R. Oostenveld, J. Onton, and S. Makeig, “Comparing results of algorithms implementing blind source separation of eeg data,” Swartz foundation and NIH grant, 2007

    2007

  35. [35]

    Independent component analysis as a tool to eliminate artifacts in eeg: a quantitative study,

    J. Iriarte, E. Urrestarazu, M. Valencia, M. Alegre, A. Malanda, C. Viteri, and J. Artieda, “Independent component analysis as a tool to eliminate artifacts in eeg: a quantitative study,”Journal of clinical neurophysiology, vol. 20, no. 4, pp. 249–257, 2003

    2003

  36. [36]

    Extraction of ocular artefacts from eeg using independent component analysis,

    R. N. Vig´ ario, “Extraction of ocular artefacts from eeg using independent component analysis,”Electroen- cephalography and clinical neurophysiology, vol. 103, no. 3, pp. 395–404, 1997

    1997

  37. [37]

    Enhanced detection of artifacts in eeg data using higher- order statistics and independent component analysis,

    A. Delorme, T. Sejnowski, and S. Makeig, “Enhanced detection of artifacts in eeg data using higher- order statistics and independent component analysis,” Neuroimage, vol. 34, no. 4, pp. 1443–1449, 2007

    2007

  38. [38]

    Independent component approach to the analysis of eeg and meg recordings,

    R. Vig´ ario, J. Sarela, V. Jousmiki, M. Hamalainen, and E. Oja, “Independent component approach to the analysis of eeg and meg recordings,”IEEE transactions on biomedical engineering, vol. 47, no. 5, pp. 589–593, 2000

    2000

  39. [39]

    Recovering eeg brain signals: Artifact suppression with wavelet enhanced independent component analysis,

    N. P. Castellanos and V. A. Makarov, “Recovering eeg brain signals: Artifact suppression with wavelet enhanced independent component analysis,”Journal of neuroscience methods, vol. 158, no. 2, pp. 300–312, 2006

    2006

  40. [40]

    Independent component analysis of electroencephalo- graphic data,

    S. Makeig, A. J. Bell, T.-P. Jung, and T. J. Sejnowski, “Independent component analysis of electroencephalo- graphic data,” inAdvances in neural information pro- cessing systems, 1996, pp. 145–151

    1996

  41. [41]

    Eeglab: an open source toolbox for analysis of single-trial eeg dynamics including independent component analysis,

    A. Delorme and S. Makeig, “Eeglab: an open source toolbox for analysis of single-trial eeg dynamics including independent component analysis,”Journal of neuroscience methods, vol. 134, no. 1, pp. 9–21, 2004

    2004

  42. [42]

    Fieldtrip: Open source software for advanced analysis of meg, eeg, and invasive electrophysiological data,

    R. Oostenveld, P. Fries, E. Maris, and J.-M. Schoffelen, “Fieldtrip: Open source software for advanced analysis of meg, eeg, and invasive electrophysiological data,” Computational Intelligence and Neuroscience, vol. 2011, p. 1, 2011

    2011

  43. [43]

    Mne software for processing meg and eeg data,

    A. Gramfort, M. Luessi, E. Larson, D. A. Engemann, D. Strohmeier, C. Brodbeck, L. Parkkonen, and M. S. H¨ am¨ al¨ ainen, “Mne software for processing meg and eeg data,”Neuroimage, vol. 86, pp. 446–460, 2014

    2014

  44. [44]

    An information- maximization approach to blind separation and blind deconvolution,

    A. J. Bell and T. J. Sejnowski, “An information- maximization approach to blind separation and blind deconvolution,”Neural computation, vol. 7, no. 6, pp. 1129–1159, 1995

    1995

  45. [45]

    A unifying information-theoretic framework for inde- pendent component analysis,

    T.-W. Lee, M. Girolami, A. J. Bell, and T. J. Sejnowski, “A unifying information-theoretic framework for inde- pendent component analysis,”Computers & Mathemat- ics with Applications, vol. 39, no. 11, pp. 1–21, 2000

    2000

  46. [46]

    Independent component anal- ysis: algorithms and applications,

    A. Hyv¨ arinen and E. Oja, “Independent component anal- ysis: algorithms and applications,”Neural networks, vol. 13, no. 4-5, pp. 411–430, 2000

    2000

  47. [47]

    Cichocki and S.-i

    A. Cichocki and S.-i. Amari,Adaptive blind signal and image processing: learning algorithms and applications. John Wiley & Sons, 2002

    2002

  48. [48]

    Robust blind source separation algorithms using cumulants,

    S. Cruces, L. Castedo, and A. Cichocki, “Robust blind source separation algorithms using cumulants,” Neurocomputing, vol. 49, no. 1-4, pp. 87–118, 2002

    2002

  49. [49]

    Amica: An adaptive mixture of independent component analyzers with shared components,

    J. A. Palmer, K. Kreutz-Delgado, and S. Makeig, “Amica: An adaptive mixture of independent component analyzers with shared components,”Swartz Center for Computatonal Neursoscience, University of California San Diego, Tech. Rep, 2012

    2012

  50. [50]

    How sensitive are eeg results to preprocessing methods: a benchmarking study,

    K. A. Robbins, J. Touryan, T. Mullen, C. Kothe, and N. Bigdely-Shamlo, “How sensitive are eeg results to preprocessing methods: a benchmarking study,” IEEE transactions on neural systems and rehabilitation engineering, vol. 28, no. 5, pp. 1081–1090, 2020

    2020

  51. [51]

    Automatic classification of artifactual ica-components for artifact removal in eeg signals,

    I. Winkler, S. Haufe, and M. Tangermann, “Automatic classification of artifactual ica-components for artifact removal in eeg signals,”Behavioral and brain functions, vol. 7, no. 1, pp. 1–15, 2011

    2011

  52. [52]

    Artifact removal techniques with signal reconstruction,

    C. A. E. Kothe and T.-P. Jung, “Artifact removal techniques with signal reconstruction,” Apr. 28 2016, uS Patent App. 14/895,440

    2016

  53. [53]

    Eeg is better left alone,

    A. Delorme, “Eeg is better left alone,”Scientific reports, vol. 13, no. 1, p. 2372, 2023

    2023

  54. [54]

    Removal of ocular artifacts using ica and adaptive filter for motor imagery-based bci,

    C. S. Kim, J. Sun, D. Liu, Q. Wang, and S. G. Paek, “Removal of ocular artifacts using ica and adaptive filter for motor imagery-based bci,”IEEE/CAA journal of automatica sinica, 2017

    2017

  55. [55]

    A comparsion of artifact rejection methods for a bci using event related potentials,

    M. Kim and S.-P. Kim, “A comparsion of artifact rejection methods for a bci using event related potentials,” in 2018 6th International Conference on Brain-Computer Interface (BCI). IEEE, 2018, pp. 1–4

    2018

  56. [56]

    On the influence of high-pass filtering on ICA-based artifact reduction in EEG-ERP,

    I. Winkler, S. Debener, K.-R. M¨ uller, and M. Tangermann, “On the influence of high-pass filtering on ICA-based artifact reduction in EEG-ERP,” in2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). IEEE, 2015, pp. 4101–4105

    2015

  57. [57]

    Deep learning for electroencephalogram (EEG) classification tasks: A review,

    A. Craik, Y. He, and J. L. Contreras-Vidal, “Deep learning for electroencephalogram (EEG) classification tasks: A review,”Journal of Neural Engineering, vol. 16, no. 3, p. 031001, 2019

    2019

  58. [58]

    Deep learning techniques for classification of electroencephalogram (eeg) motor imagery (mi) signals: a review,

    H. Altaheri, G. Muhammad, M. Alsulaiman, S. U. Amin, G. A. Altuwaijri, W. Abdul, M. A. Bencherif, and M. Faisal, “Deep learning techniques for classification of electroencephalogram (eeg) motor imagery (mi) signals: a review,”Neural Computing and Applications, pp. 1– 42, 2021

    2021

  59. [59]

    Inter-subject transfer learning with an end-to-end deep convolutional neural network for EEG- based ,

    F. Fahimi, Z. Zhang, W. B. Goh, T.-S. Lee, K. K. Ang, and C. Guan, “Inter-subject transfer learning with an end-to-end deep convolutional neural network for EEG- based ,”Journal of neural engineering, vol. 16, no. 2, p. 026007, 2019

    2019

  60. [60]

    Learning eeg topographical representation for classification via convolutional neural network,

    M. Xu, J. Yao, Z. Zhang, R. Li, B. Yang, C. Li, J. Li, and J. Zhang, “Learning eeg topographical representation for classification via convolutional neural network,” Pattern Recognition, vol. 105, p. 107390, 2020. I See Artifacts: ICA-based EEG Artifact Removal Does Not Improve Deep Network Decoding Across Three BCI Tasks23

    2020

  61. [61]

    Mi-eegnet: A novel convolutional neural network for motor imagery classification,

    M. Riyad, M. Khalil, and A. Adib, “Mi-eegnet: A novel convolutional neural network for motor imagery classification,”Journal of Neuroscience Methods, vol. 353, p. 109037, 2021

    2021

  62. [62]

    Op- timizing motor intention detection with deep learn- ing: towards management of intraoperative awareness,

    O. Avilov, S. Rimbert, A. Popov, and L. Bougrain, “Op- timizing motor intention detection with deep learn- ing: towards management of intraoperative awareness,” IEEE Transactions on Biomedical Engineering, vol. 68, no. 10, pp. 3087–3097, 2021

    2021

  63. [63]

    Multiclass classifica- tion of spatially filtered motor imagery eeg signals us- ing convolutional neural network for bci based applica- tions,

    N. Shajil, S. Mohan, P. Srinivasan, J. Arivudaiyanambi, and A. Arasappan Murrugesan, “Multiclass classifica- tion of spatially filtered motor imagery eeg signals us- ing convolutional neural network for bci based applica- tions,”Journal of Medical and Biological Engineering, vol. 40, no. 5, pp. 663–672, 2020

    2020

  64. [64]

    Study on the effect of different electrode channel combinations of motor imagery eeg signals on classification accuracy,

    K. Zhu, S. Wang, D. Zheng, and M. Dai, “Study on the effect of different electrode channel combinations of motor imagery eeg signals on classification accuracy,” The Journal of Engineering, vol. 2019, no. 23, pp. 8641– 8645, 2019

    2019

  65. [65]

    Mu rhythm (de) synchronization and eeg single-trial classification of different motor imagery tasks,

    G. Pfurtscheller, C. Brunner, A. Schl¨ ogl, and F. L. Da Silva, “Mu rhythm (de) synchronization and eeg single-trial classification of different motor imagery tasks,”NeuroImage, vol. 31, no. 1, pp. 153–159, 2006

    2006

  66. [66]

    Iterative gaussianization: from ica to random rotations,

    V. Laparra, G. Camps-Valls, and J. Malo, “Iterative gaussianization: from ica to random rotations,”IEEE transactions on neural networks, vol. 22, no. 4, pp. 537– 549, 2011

    2011

  67. [67]

    Eeg- based prediction of successful memory formation during vocabulary learning,

    T. Kang, Y. Chen, S. Fazli, and C. Wallraven, “Eeg- based prediction of successful memory formation during vocabulary learning,”IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 28, no. 11, pp. 2377–2389, 2020

    2020

  68. [68]

    Eegmanypipelines: A large-scale, grass-root multi-analyst study of eeg analysis practices in the wild,

    D. Tr¨ ubutschek, Y.-F. Yang, C. Gianelli, E. Cesnaite, N. L. Fischer, M. C. Vinding, T. Marshall, J. Algermissen, A. Pascarella, T. Puoliv¨ aliet al., “Eegmanypipelines: A large-scale, grass-root multi-analyst study of eeg analysis practices in the wild,” 2022

    2022

  69. [69]

    Blind source separation,

    G. R. Naik, W. Wanget al., “Blind source separation,” Berlin: Springer, vol. 10, pp. 978–3, 2014

    2014

  70. [70]

    Heartbeat classification using morphological and dynamic features of ecg signals,

    C. Ye, B. V. Kumar, and M. T. Coimbra, “Heartbeat classification using morphological and dynamic features of ecg signals,”IEEE Transactions on Biomedical Engineering, vol. 59, no. 10, pp. 2930–2941, 2012

    2012

  71. [71]

    Application of independent component analysis with adaptive density model to complex- valued fmri data,

    H. Li, N. M. Correa, P. A. Rodriguez, V. D. Calhoun, and T. Adali, “Application of independent component analysis with adaptive density model to complex- valued fmri data,”IEEE Transactions on biomedical engineering, vol. 58, no. 10, pp. 2794–2803, 2011

    2011

  72. [72]

    Temporally constrained ica: an application to artifact rejection in electromag- netic brain signal analysis,

    C. J. James and O. J. Gibson, “Temporally constrained ica: an application to artifact rejection in electromag- netic brain signal analysis,”IEEE Transactions on biomedical engineering, vol. 50, no. 9, pp. 1108–1116, 2003

    2003

  73. [73]

    Removing electroencephalographic artifacts by blind source separation,

    T.-P. Jung, S. Makeig, C. Humphries, T.-W. Lee, M. J. Mckeown, V. Iragui, and T. J. Sejnowski, “Removing electroencephalographic artifacts by blind source separation,”Psychophysiology, vol. 37, no. 2, pp. 163–178, 2000

    2000

  74. [74]

    Blind source separation, wavelet denoising and discriminant analysis for eeg artefacts and noise cancelling,

    R. R. V´ azquez, H. Velez-Perez, R. Ranta, V. L. Dorr, D. Maquin, and L. Maillard, “Blind source separation, wavelet denoising and discriminant analysis for eeg artefacts and noise cancelling,”Biomedical signal processing and control, vol. 7, no. 4, pp. 389–400, 2012

    2012

  75. [75]

    An overview of independent component analysis and its applications,

    G. R. Naik and D. K. Kumar, “An overview of independent component analysis and its applications,”Informatica, vol. 35, no. 1, 2011

    2011

  76. [76]

    Ica-based eeg denoising: a comparative analysis of fifteen methods,

    L. Albera, A. Kachenoura, P. Comon, A. Karfoul, F. Wendling, L. Senhadji, and I. Merlet, “Ica-based eeg denoising: a comparative analysis of fifteen methods,” Bulletin of the Polish Academy of Sciences: Technical Sciences, vol. 60, no. 3 Special issue on Data Mining in Bioengineering, pp. 407–418, 2012

    2012

  77. [77]

    Independent eeg sources are dipolar,

    A. Delorme, J. Palmer, J. Onton, R. Oostenveld, and S. Makeig, “Independent eeg sources are dipolar,”PloS one, vol. 7, no. 2, p. e30135, 2012

    2012

  78. [78]

    A blind source separation technique using second-order statistics,

    A. Belouchrani, K. Abed-Meraim, J.-F. Cardoso, and E. Moulines, “A blind source separation technique using second-order statistics,”IEEE Transactions on signal processing, vol. 45, no. 2, pp. 434–444, 1997

    1997

  79. [79]

    Blind beamforming for non-gaussian signals,

    J.-F. Cardoso and A. Souloumiac, “Blind beamforming for non-gaussian signals,” inIEE proceedings F (radar and signal processing), vol. 140, no. 6. IET, 1993, pp. 362– 370

    1993

  80. [80]

    Modeling and estimation of dependent subspaces with non-radially symmetric and skewed densities,

    J. A. Palmer, K. Kreutz-Delgado, B. D. Rao, and S. Makeig, “Modeling and estimation of dependent subspaces with non-radially symmetric and skewed densities,” inIndependent Component Analysis and Signal Separation: 7th International Conference, ICA 2007, London, UK, September 9-12, 2007. Proceedings

    2007

Showing first 80 references.