睡眠ステージの知識データベースに基づく条件付確立を用いた自動判定法

摘要

Sleep consists of non-rapid eye movement (NREM) and rapid eye movement (REM) states. NREM is further subdivided into Stage I, II, III and IV. The most well-known criteria for sleep stage scoring were published by Rechtschaffen and Kales in 1968. Each state is characterized by a different type of brain wave activity. Currently, sleep stage scoring has been widely used for evaluating the condition of sleep and diagnosing the sleep related disorders in hospitals and institutions. Automatic sleep stage determination can free the clinicians from the heavy task of visual inspection on sleep stages. Rule-based waveform detection methods, according to Rechtschaffen and Kales criteria, have been developed in many studies. However, Rechtschaffen and Kales criteria including typical waveforms of healthy persons under ideal condition for sleep stage scoring are insufficient to cover the variable sleep data of patients under usual condition in hospitals. The conventional rule-base methods have the similar limitations for clinical practice. An expert knowledge-based probabilistic method is developed in order to overcome the limitation of conventional rule-based methods. The visual inspection of sleep stage scoring by a qualified clinician is adopted as the expert knowledge. According to the visual inspection on a set of training data, an expert knowledge database is established in terms of probability density functions of parameters for various sleep stages. A set of characteristic parameters are defined as candidates. The probability density functions for various sleep stages are developed by using Cauchy distribution to approximately estimate the parameter distribution on histogram. The parameter which is effective for sleep stage discrimination is selected automatically. Sleep stages is determined automatically by the maximum value of conditional probabilities. An amendment function is developed to modify the decision making of sleep stage by the expert knowledge-based method, which is designed according to the additional rules by clinician for the continuity of stage II and onset/offset of stage REM. The developed expert knowledge-based automatic sleep stage determination system has flexible performance for clinical practice.