Feature Extracting of Weak Signal in Real-Time Sleeping EEG with Approximate Entropy and Bispectrum Analysis

Electroencephalogram (EEG) is one of the most important neuroelectrical signals and is often used to detect brain's neuroelectrical dysfunction. However, the analysis of the EEG signal and extraction of features from it has been a challenging task due to the complexity and variability. It is difficult to recognize different stages of the real-time sleeping EEG signal from single EEG signal automatically. In our study, the features were extracted from single sleeping EEG signal of rats using approximate entropy (ApEn) combined with bispectrum analysis. The results show that ApEn and the maximal amplitude of bispectrum can extract various features in different sleeping EEG stages. The bispectrum can detect the phase coupling among different stages in sleeping EEG signal. The ApEn with change of bispectrum's frequency and maximal amplitude of bispectrum can be effectively applied to automatically compartmentalize real-time sleeping EEG signal into different stages, which is significant to automatic EEG analyze and intelligent diagnose of brain diseases. The results also provide a new way of features extraction for other non-stationary signals in real time.