Background: Conventional manual sleep stage classiﬁcation is time-consuming and relies on the knowledge and experience of the specialists. The emergence of automatic sleep stage classiﬁcation greatly improves the classiﬁcation eﬃciency. The feature extraction in automatic sleep stage classiﬁcation is particularly important, which usually uses the linear methods based on techniques in the time domain, frequency domain, or time-frequency domain. Electroencephalograms (EEGs) contain a wealth of physiological information, have been widely used for the classiﬁcation of sleep stage. Due to the nonlinear, non-stationary, and multifractal characteristics of EEGs, some nonlinear methods have been used to extract features of sleep stages in recent years, such as complexity, multifractal theory, and chaos theory. The Wavelet Leader Multifractal Formalism (WLMF) of the multifractal theory is widely applied to diﬀerent physiological signals. The current researches focus on discussing the mean H¨older exponent (h0) and the width of the multifractal singularity spectrum (WD(h)) estimated by the WLMF method. However, in the ﬁeld of sleep staging, a number of researches focused on h0, but few studies on WD(h).
Results: This paper aims to assess the multifractal characteristic for sleep EEG time series from the Sleep-EDF Expanded Database by the WLMF method. In the young group, the mean h0 increased from the Wake stage to the S3 stage (p<0.01). So did the elderly group (p<0.001). WD(h) of the Wake stage was less than that of the S3 stage for the young group, and this relationship was reversed for the elderly group(χ2=13.769, df=1, p<0.001). Gender did not aﬀect, with statistical signiﬁcance, WD(h) of the Wake stage and the S3 stage (χ2=0.085, df=1, p=0.608), nor did the brain region (χ2=3.137, df=1, p=0.078).
Conclusions: The result shows that WD(h) was inﬂuenced by aging. The gender and location of brain regions did not show signiﬁcant inﬂuence on the multifractal characteristics of wakefulness and sleeping. This ﬁnding extends the application of the multifractal singularity spectrum on sleep staging, and raises a fundamental question on what might be the underlying mechanisms of the WD(h) reversion.