Wavelet-based segmentation techniques in the detection of microarousals in the sleep EEG.

摘要

This thesis proposes an automatic detection procedure to detect the presence of undesirable frequency bursts, called microarousals (MA), within any of the various stages of sleep. Sleep is examined through the acquisition of the electroencephalogram (EEG). Traditionally, a sleep technologist manually inspects the EEG signal to correctly detect the occurrence of MAs. The presence of these MAs causes a medical condition known as excessive daytime sleepiness (EDS).; Since the EEG is a non-stationary signal, the proposed procedure analyzes it in three stages. The first stage involves spectral decomposition using the discrete wavelet transform (DWT). The DWT is efficient and possesses excellent time-frequency resolution that makes it well suited to exploit the characteristics of a non-stationary signal.; The second stage of the proposed procedure partitions the decomposed signal into stationary segments. Both parametric and nonparametric segmentation techniques are applied. The nonparametric autocorrelation function (ACF) and the nonlinear energy operator (NLEO) methods as well as the parametric generalized likelihood ratio (GLR) method are each applied to the component waveforms of the EEG signal produced by the DWT.; The third stage of the proposed procedure involves evaluating information about each stationary segment's power and spectral content. Once this information is determined, segments satisfying the definition of a MA are detected and scored.; To examine the effectiveness of the overall procedure, long-term EEG records containing MAs that have been marked by a sleep technologist are compared against the proposed procedure's detected MAs. The successful results obtained demonstrate the effectiveness of the proposed procedure.