Nonlinear Dimensionality Reduction and Feature Analysis for Artifact Component Identification in hdEEG Datasets
The time-domain independent component analysis (ICA) is commonly used technique for neural and noise signal separation in electroencephalographic (EEG) data. Nevertheless, the estimated independent sources have to be further classified by an expert using predefined or learned features. Automatic algorithms on market suffer from unsatisfactory sensitivity and specificity mostly due to large inter-subject variability of the human EEG. Generalization of each learning machine depends on whether some discriminative features exist in inter and intra individual component space. We employ nonlinear dimension reduction technique called t-Distributed Stochastic Neighbor Embedding (t-SNE) to fit the data and visualize them in estimated feature space. Our approach is unsupervised and completely data driven. We focus on regularities in a real-world hdEEG component set that could be learned by robust feature classifier, e.g. a deep network. Furthermore, we also investigate relations between found data structure and commonly used criteria known from the literature.
KeywordsIndependent component analysis High-density EEG Artifact Dimension reduction t-Distributed stochastic neighbor embedding DBSCAN
This work was supported by the Grant Agency of the Czech Technical University in Prague, registr. numb. SGS18/159/OHK4/2T/17 with topic: Feature space analysis using linear and nonlinear reduction of EEG space dimensions, by the Grant Agency of Czech Republic with topic: Temporal context in analysis of long-term non-stationary multidimensional signal, register number 17-20480S, by the grant AZV 15-29370A, and project LO1611 with a financial support from the MEYS under the NPU I program.
Statements by Authors
The authors declare that there is no conflict of interest regarding the publication of this article.
The study protocol and patient informed consent have been approved by National Institute of Mental Health ethical committee.
The procedures followed were in compliance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the World Medical Association Declaration of Helsinki on Ethical Principles for Medical Research Involving Human Subjects.
- 1.M. Chaumon, D.V. Bishop, N.A. Busch, Journal of neuroscience methods 250, 47 (2015).Google Scholar
- 2.H. Nolan, R. Whelan, R.B. Reilly, Journal of Neuroscience Methods 192(1), 152 (2010). https://doi.org/10.1016/j.jneumeth.2010.07.015.
- 3.A. Mognon, J. Jovicich, L. Bruzzone, M. Buiatti, Psychophysiology 48(2), 229 (2011). https://doi.org/10.1111/j.1469-8986.2010.01061.x.
- 4.I. Winkler, S. Brandl, F. Horn, E. Waldburger, C. Allefeld, M. Tangermann, Journal of Neural Engineering 11(3) (2014). https://doi.org/10.1088/1741-2560/11/3/035013.
- 5.F. Campos Viola, J. Thorne, B. Edmonds, T. Schneider, T. Eichele, S. Debener, Clinical Neurophysiology 120(5), 868 (2009). https://doi.org/10.1016/j.clinph.2009.01.015.
- 6.T. Radüntz, J. Scouten, O. Hochmuth, B. Meffert, Journal of Neural Engineering 14(4), 046004 (2017). https://doi.org/10.1088/1741-2552/aa69d1.
- 7.R.T. Schirrmeister, L. Gemein, K. Eggensperger, F. Hutter, T. Ball, arxiv:1708.08012 (2017).
- 8.L.V.D. Maaten, G. Hinton, Journal of Machine Learning Research 9, 2579 (2008). https://doi.org/10.1007/s10479-011-0841-3.
- 9.R. Oostenveld, P. Fries, E. Maris, J.M. Schoffelen, Computational intelligence and neuroscience 2011, 1 (2011).Google Scholar
- 10.T.W. Lee, M. Girolami, T.J. Sejnowski, Neural computation 11(2), 417 (1999).Google Scholar
- 11.M. Ester, H.P. Kriegel, J. Sander, X. Xu, et al., in Kdd, vol. 96 (1996), pp. 226–231.Google Scholar