Automated detection of low-dimensional EEG alpha-episodes. An example of application to psychopharmacological data

This paper addresses the practical problems raised by the fact that correlation integrals computed from electroencephalographic signals do not scale. An automated procedure is described for alpha-waves, which is based on the known method of determining the correlation dimension from scaled structures that appear in diagrams of slope-curves. Automation, which is achieved using a moving window that detects the minima of slope-curves, and the density of minima makes it possible to vary the numerous parameters, and to implement tests. The potentialities of the automatic procedure are illustrated by results of psychopharmacological interest, based on the analysis of 22x10(3) alpha-signals, and the computation of 32x10(6) correlation integrals. Previous results concerning the existence of low-dimensional alpha-episodes, lasting in general not more than 5-6 s, are confirmed. Opposing effects upon the correlation dimension are described, at their pharmacological peak, for a dopaminergic agonist (apomorphine) and an atypical antipsychotic, a sigma-receptor ligand. The results also confirm that time reparametrization is a vital check in searches of low-dimensional attractor-ruled dynamics, and they suggest that subjects exhibiting 'high-alpha', i.e. whose alpha-waves are of amplitude A>30 microV, be selected for such investigations.