Sleep stage diagnosis system with neural network analysis

Information of sleep stage is one of the most important clue for diagnosis of mental condition and disease in psychiatry. So far we have proposed a method of detection of characteristic waves in sleep EEG and diagnosing the sleep stages of the segmented short-terms by neural networks analysis. And we showed that it was able to diagnose the sleep stages to some extent by recognizing the time-varying spectral patterns of characteristic waves. There remains, however, a problem that results in stage diagnosis often become unstable, since the contextual relation between the present and adjacent segments is not considered. In this work, a method of diagnosing the sleep stage more accurately is proposed and its performance is evaluated. In the method, the additional neural networks processing is combined with the previous system for recognizing the context of stage sequences. As a result, it is proved that detection ratio is improved to a considerable extent by utilizing the contextual information on stages and the proper duration exists for obtaining high performance.