Psychophysiology, 37:S54, 2000.
Principal components analysis (PCA) as a tool for identifying EEG frequency bands: I. Methodological considerations and preliminary findings
Jürgen Kaysera, Craig E. Tenkea, Stefan Debenerb
a Department of Biopsychology, New York State Psychiatric Institute, New York, NY 10032, USA
b Dresden University of Technology, Department of Psychology II, Germany
Abstract
Topographies of quantitative EEG (qEEG) measures, particularly alpha (8-13 Hz) power, are commonly used in clinical and basic research. However, inferences about brain activation derived from somewhat arbitrarily defined EEG frequency bands are often hampered by variability in the spectra across groups and experimental condition. Moreover, qEEG topographies may vary within a frequency band, and the exact number and range of possible subbands (e.g., low and high alpha) is also unclear. This problem is precisely analogous to the identification of ERP components, but in this case, the waveform is a positive-valued function of frequency. While an ERP diverges from zero during the poststimulus period, EEG spectra converge to zero at high frequencies. Condition-related (resting eyes open/closed) variations are likewise small for high frequencies, but large at or below alpha. The merit of covariance-based PCA (unscaled Varimax rotation) for identifying overlapping ERP components let us hypothesize that this particular methodology would also determine systematic variations of frequency amplitudes, eventually yielding more useful frequency components (i.e., "data-driven" frequency bands). PCAs performed on 30-channel resting EEG data (nose reference) reliably extracted factors matching different frequency ranges, among them four distinct factors that all included a: 1) loading peak between 8 and 13 Hz; 2) condition effect; 3) posterior topography. Future studies need to investigate the usefulness of these factors, and the importance of the frequency resolution for factor extraction.
Keywords: EEG methodology; frequency PCA; alpha