Psychophysiology, 40:S84, 2003.

Identification and Separation of Reference-free Spectral EEG Components: Combining Current Source Density (CSD) and Frequency Principal Components Analysis (fPCA)

C.E. Tenke and J. Kayser

Department of Biopsychology, New York State Psychiatric Institute, New York, NY 10032, USA

Abstract

The interpretability of quantitative EEG is severely limited by two crucial choices: definition of appropriate frequency bands and recording reference. Both the spectral properties of the EEG and their associated topographies, as well as those of artifacts, can vary considerably for any given subject and/or paradigm, based on these choices. Some researchers have acknowledged these problems by simultaneously analyzing various reference schemes, or by more loosely defining frequency bands to match data properties (e.g., varying alpha ranges to capture a peak). Problems are exacerbated when power spectra are computed (e.g., information is lost by nonlinear transformation, hemispheric asymmetries may invert, etc). In contrast, a surface Laplacian (CSD) is a data transformation with known correspondence to neuronal generators. CSDs concisely reflect the true topographic variation of the EEG signal, ignore volume-conducted activity, and are identical for any reference scheme (e.g., linked ears/mastoids, nose, average, etc). A fPCA provides a concise summary of EEG spectra that conforms to the data, rather than to rigid, unrelated frequency bands. Combining these advantages, amplitude spectra were computed from CSD epochs (eyes open/closed 30-channel resting EEG of 72 subjects), and summarized using an unrestricted, Varimax-rotated, covariance-based fPCA. Multiple, psychophysiologically meaningful alpha factors with distinct topographies were clearly separable from eye and muscle artifacts, providing a more relevant topographic description of the spectral structure of these data.

Keywords: EEG methodology, surface Laplacian, frequency PCA