Methodological Framework for EEG Feature Selection Based on Spectral and Temporal Profiles
Among the various frameworks in which EEG signal analysis has been traditionally formulated, the most widely studied is employing power spectrum measures as functions of certain brain pathologies or increased cerebral engagement. Such measures may form signal features capable of characterizing and differentiating the underlying neural activity. The objective of this chapter is to validate the use of wavelets in extracting such features in the time–scale domain and evaluate them in a simulated environment assuming two tasks (control and target) that resemble widely used scenarios of assessing and quantifying complex cognitive functions or pathologies. The motivation for this work stems from the ability of time–frequency features to encapsulate significant power alteration of EEG in time, thus characterizing the brain response in terms of both spectral and temporal activation. In the presented algorithmic scenario, brain areas’ electrodes of significant activation during the target task are extracted using time-averaged wavelet power spectrum estimation. Then, a refinement step makes use of statistical significance-based criteria for comparing wavelet power spectra between the target task and the control condition. The results indicate the ability of the proposed methodological framework to correctly identify and select the most prominent channels in terms of “activity encapsulation,” which are thought to be the most significant ones.
KeywordsWavelet Transform Morlet Wavelet Target Task Wavelet Power Spectrum Feature Selection Step
This work was supported in part by the EU IST project BIOPATTERN, Contact No: 508803. The Wavelet Transform and various significance testing parts were performed using software implementation based on the wavelet toolbox provided by C. Torrence and G. Compo, available at the URL: http://paos.colorado.edu/ research/ wavelets/.
- 1.Bianchi, A., Mainardi, L., Cerutti, S. Time-frequency analysis of biomedical signals. Trans Inst Measur. and Contr 22, 321–336 (2000)Google Scholar
- 3.Bressler, S. The Handbook of Brain Theory and Neural Networks. MIT Press, Cambridge MA 412–415, (2002)Google Scholar
- 5.Burrus, S., Gopinoth, R., Haitao, G. Introduction to Wavelets and Wavelet Transforms. Prentice Hall Inc., Upper Saddle River, NJ (1998)Google Scholar
- 6.Chatfield, C. The Analysis of Time Series: An Introduction, 4th edn, p. 241. Chapman and Hall, London (1989)Google Scholar
- 12.Raudys, S., Pikelis, V. On dimensionality, sample size, classification error and complexity of classification algorithm in pattern recognition. IEEE Trans Pattern Anal Machine Intell, PAMI-2(3), 242–252 (1980)Google Scholar
- 13.Sakkalis, V., Giurcaneanu, C., Xanthopoulos, P., Zervakis, M., Tsiaras, V., Yang, Y., Micheloyannis, S. Assessment of linear and nonlinear synchronization measures for analyzing EEG in a mild epileptic paradigm. Accepted to be published in, 2008. IEEE Trans Inf Tech (2008). DOI: 10.1109/TITB.2008.923141.Google Scholar
- 16.Stoica, P., Moses, R. Introduction to Spectral Analysis. Prentice-Hall, Upper Saddle River, NJ (1997). 24–26Google Scholar
- 19.Zar, J.: Biostatistical Analysis, 4th edn. Prentice Hall, Upper Saddle River, NJ (1999)Google Scholar