Effects of propofol anesthesia on nonlinear properties of EEG: Lyapunov exponents and short-term predictability

This paper focuses on the analysis of the short term predictability of the EEG signal after the estimation of the maximum Lyapunov exponent (MLE) of the chaotic attractor associated to the EEG signal. After clinical research and ethical committee approval and signed informed consent, EEG data was collected from 6 patients scheduled for surgery under general anesthesia and 7 ones scheduled for ambulatory endoscopic procedures. One differential channel of EEG (+ in mastoids M1/M2, - in the middle-line of the forehead FPz/AFz and referenced to F7/F8) was amplified with a gain of 10.000 (Biopac MP100-EEG100B), low-pass filtered at 300 Hz (2nd order Butterwoth) and acquired at 2.5 KHz. Data were digitally filtered (50 Hz comb filter and 120-taps linear phase FIR lowpass filter at 85 Hz) and then decimated to 250 Hz prior to analysis under OpenTstool. In order to reconstruct the associated chaotic attractor, two different approaches were followed. First, classical methods were used to find time-lag and embedding dimension. Also, a parametric swept was used for tau=1,2,3...,20 and d=2,3...6. Once the attractor was reconstructed, MLE was estimated as the finite growth rate proposed by Wessel et al., for different window lengths. Surrogate data sets were generated for validation purposes. After analyzing the data with different window lengths (250, 500, 750, 1000, 1500 and 2000 samples), finite growth rates were found to decrease with anesthesia. Embedding after formal methods did NOT offer the best discrimination among different anesthesia levels. Discrimination was very effective even with very short time series (250 samples - 1s). These results seem to indicate that finite growth rates are able of detecting the changes on brain dynamics induced by anaesthetic agents. © 2009 Springer Berlin Heidelberg.