Unsupervised classification and adaptive definition of sleep patterns

The classical criteria for definition of the various sleep stages, formulated by Rechtschaffen and Kales, do not hold for automatic sleep scoring of polygraphic sleep recordings in sleep disturbances and for sleep affected by CNS drugs. Classification of the sleep EEG into the basic sleep patterns (sleep stages) is carried out using first adaptive segmentation, as a phase of feature selection, and then fuzzy clustering. The optimal number and character of the basic sleep patterns is considered a priori unknown, and is estimated by calculating fuzzy performance measures. This method of unsupervised classification and adaptive definition of the various sleep stages has been tested on sleep recordings of normal subjects and on patien ts with sleep disturbances and under CNS medication. The optimal number of sleep stages, and the prototype for each stage have been computed for each of the sleep recordings, generating a subject-specific sleep print.