Characteristics Analysis Of The Eeg And Sleep Stage Study

The sleep is a kind of important physiological phenomenon. In psychiatry, sleep staging is one of the most important means for diagnosis. And the electroencephalogram (EEG) is a very important tool foranalyzingthe sleep because a large amount of physiology and pathology information in EEG signals. It plays a very important role in the field of clinical medicine and brain science. In this paper, wavelet transform and Independent Component Analysis (ICA) are used to process the EEG signals, including EEG de-noising and artifact removing. Further, power spectrum analysis and complexity measures are calculated.The EEG signals are usually very weak, so that they are easy to be disturbed by noises. The noises are composed of ECG, EMG, EOG, impulse signal and high-frequency noise. The EOG artifact is removed from the EEG with ICA because of the mutual independence between them. To the high-frequency noise, it is very difficult to cancel it in the time domain because the noise exists in the whole frequency band of EEG signal. We can get decomposing result in different scales with the wavelet transform. Because the noise is concentrated in high frequency domain, by filtering the high frequency noise and then reconstructing the signal, we can filter the signal. We use a threshold de-noising algorithm which can achieve a very good filtering effect.When a person performs different mental tasks, the EEG signals change a lot. If the feature of different mental EEG can be detected, then a new method can be developed to study BCI technology. And it is significant for knowing the process of thoughts and the principle of brain. In this paper we compute the Approximate Entropy (ApEn) of five mental tasks for one person. And the ApEn is used to analyze the mental EEG signals for the pattern extraction and mental task classification. Based on the ApEn changes of different mental tasks, a neural network is trained to perform the classification. The results show that classification accuracy is up to 80%.For a whole night recordings, there will be a great deal of sleep EEG data. Human sleep stages are hand-scored by an expert is a laborious, time consuming and difficult task. To overcome these difficulties, developing some automatic sleep stages is significant. In this paper, we calculate the ApEn of sleep EEG signals and determine the sleep stages based on these ApEn parameters. ApEn is the biggest in wake stage. From wake stage to stage I and II the ApEn become smaller, and in stage III and IV ApEn is the smallest. But in REM stage ApEn will go up to the value near to stage I . All ApEn of sleep stages are different for each other except for the stage III and IV which is difficult to distinguish. Comparing with the results of expert's analysis, the mean rate of accordance is up to 75%. For the results of stage III and IV, we can utilize the AR model to distinguish them. The AR coefficients can be regarded as the characteristic coefficients of the EEG signals and inputted into a BP neural network to distinguish the stage III and IV, and its rate of accordance is up to 70%. In conclusion, the method we provide can be used to analyze and process the sleep EEG signals. It also can be used for the automatic sleep staging. The result shows that the performance of the method is satisfying.