Differentiation of normal and disturbed sleep by automatic analysis.

The main purpose of the present study was to develop an automatic hybrid system and evaluate its performance in the differentiation of normal and disturbed sleep. The study was carried out in roughly the following steps: At the beginning the quantity of five EEG waveforms, delta, theta, alpha, sigma and beta were examined with respect to sleep stages and their temporal distributions and inter-relationships throughout the nights in young 20-22 year-old, healthy individuals. These data were used as a basis for the choice of parameters for a sleep stage scoring program. Secondly, the sleep stage scoring software was developed using as test material three subject groups: about 10 years older controls, anonymous alcoholics and chronic alcoholics in the withdrawal phase. The last group was examined on two occasions, first in the initial withdrawal and then after two weeks' abstinence between the recordings. The sleep stage scoring program was evaluated in comparison with a visual sleep stage determination and its performance was also tested on the young normals. Attention was paid to three main points: First, how well does an automatic system imitate the visual classification of a human subject, when both the human and the computer classifications are compared side by side at a 20 s epoch level? Second, even if this comparison did not show a satisfactory agreement, could the computer classification still replace the visual sleep stage scoring when the objective is to describe the neurophysiological characteristics of the human whole-night sleep process or differentiate between normal and disturbed sleep. For this purpose it was examined, whether the values for the sleep stage parameters obtained by visual classification corresponded to those obtained by computer scoring and whether the differences found between the groups by manual methods could also be obtained by computer classification. The third goal was to see whether other measures than parameters received by sleep stage classification could be used for the differentiation between normal and disturbed sleep. In the present work only EEG waveform parameters and body movement activity were studied with this in mind. It was found that sleep can satisfactorily be classified in stages by automatic analysis if it is not markedly disturbed. The percentage agreement obtained for the three groups having practically normal sleep (young normals appr. 80%, older normals 77% and anonymous alcoholics 75%) was satisfactory and sufficient for clinical and experimental work.(ABSTRACT TRUNCATED AT 400 WORDS)