Analysis of Sleep Staging Based on Multivariate Symbolic Transfer Entropy

Abstract

Physiological electrical signals such as ECG, EEG and EMG during sleep contain a lot of physiological information. The analysis of these signals can provide effective advice for the diagnosis of sleep staging or sleep disorders. The symbol transfer entropy algorithm is applied to study the interaction information of physiological signals. However, because the traditional symbol transfer entropy algorithm only focuses on the relationship between one or two variables, this paper used the multivariate symbolic transfer entropy based on the traditional symbol transfer entropy to consider the coupling relationship between multiple variables. When dividing the time series, we used two methods respectively: static partitioning and dynamic adaptive. By comparing the multivariate symbolic transfer entropy values in the awake and sleep periods of different subjects, we can get that the multivariate symbol shift entropy of the awake period is significantly higher than that of the sleep subjects, although the entropy values are different from those of the subjects. And there is a significant difference between the two by using T test, which is consistent with the theory that the degree of brain disorder decreases and the entropy decreases when the sleep is deepen. The multivariate symbolic transfer entropy algorithm is effective in distinguishing human awake period from sleep period and can provide an effective way for other physiological signals research.