Repeated characterization, analysis of variance limitation and discrimination analysis classification of EEG-activity patterns in human sleep]

1. To describe quantitatively and to deliminate nine EEG sleep patterns, mean values and standard deviations of abundances of the frequencies 0.8 ... 1.8 c/sec, 2...3.5 c/sec, 4...13c/sec, 14 to 17 c/sec, 18 to 22 c/sec, and 23 to 40 c/sec as well as of the average amplitudes in selected frequency ranges were calaculated and the distributions represented. 2. All nine EEG activity patterns could be separated by means of univariate and multivariate analyses of variance on the basis of all 28 as well as the 17 indispensable variables. 3. In the course of a stepwise reduction of variables within the framework of a linear discriminant analysis an optimal set of 17 variables was determined for the separation of the patterns, comprising: the percent quantity of the frequencies 0.8 ... 3.5 c/sec, 7 ... 9 c/sec and 18 to 40 c/sec as well as the average amplitudes in the frequency ranges 0.8 to 3.5 c/sec and 7.5 to 40 c/sec. 4. By linear regression analyses it could be shown that the sleep scording system used, can be reflected on an interval scale with the aid of discriminant functions; this can be achieved on the basis of the optimal set of variables as well as of the five most indispensable variables. 5. Finally the degree of the objectivity of the scoring procedures was demonstrated. Advantages and disadvantages of sleep scoring systems were discussed and possibilities of the utilization of results suggested, also in respect to the further development of the automatic recognition of EEG activity patterns.