A multivariate approach to the analysis of drug effects on the electroencephalogram.

The present approach was developed to achieve a continuous analysis of the EEG over periods of several hours. An analogue frequency analyzer is used spanning the range 2.5-40 Hz in eight half-octave bands. At one-minute intervals the rectified and integrated channel outputs are digitized. Four brain areas have been used in most experiments yielding values of 32 variables per minute. Discriminant analysis allows the classification of behavioral state (slow sleep, fast sleep, or alert) with >99% accurary in cats. Canonical analysis shows that spontaneous variations in EEG frequency distribution may be described adequately by only two orthogonal variables. Drug-induced changes may be described as projections on the plane thus defined and/or in terms of a third, drug-specific canonical variable. The value of this variable estimates the intensity of the drug-induced change; a profile of frequency change specific to the drug effect may be derived from its coefficients.