Robust Spectral Analysis of the EEG

Abstract

Robust methods for the spectral analysis of time series are briefly reviewed and seen to have applications in the field of EEG. After presenting two simple schemes for outliers (artifacts) generation and discussing their implications for estimation of the spectral density, the robust filtering algorithm of Kleiner et al. [J. R. Statist. Soc. Ser. B, 41: 313-351, 1979] is introduced and shown to work well for simulated data and for true EEG data containing artifacts. A new use of the robust methods for the detection of artifacts and possibly other transients in long-time EEG recordings is suggested and a preliminary implementation illustrated.

Spectral Analysis of the EEG

“…From an abstract statis tical point of view, the requirements, in terms of stability and bias and/or resolution, of the estimates to be com puted for the problem at hand should be first formu lated; then, when worked out in detail, the asymptotics given by Dumermuth and Molinari [1986] provide (pos sibly after a preliminary study to determine data-dependent parameters of the estimation procedure, and if the necessary care has been used, e.g. to avoid artifacts) an approximation of the required record length.…”

“…3) of applying this proce dure to two channels of EEG data, where some emphasis is on the robust estimation of coherence. Additional examples as well as more detailed considerations about robustness in time series are given in Molinari and Dumermuth [1986].…”

Oscillation features time series and linear - nonlinear dynamic regression models with application in predicting epileptic seizure

“…see Ille et al (2002); Delorme and Makeig (2004)), as well as methods for EEG analysis that are robust to the existence of such artifacts (e.g. see Molinari L (1986)). In general however, explicit detection and correction of artifacts is not an easy task to implement as there is a very wide variation of artifacts, some of which cannot be distinguished from normal brain activity.…”