The Extraction And Analysis Of EEG Features

An electroencephalogram (EEG) record brainwaves patterns from the continuous rhythm and spontaneous electrical signals coming from the brain via electrodes, which contains a lot of information about physiology and pathology. EEG helps to improve the reliability and accuracy in clinical diagnosis and detection about the brain neural pathological changes. As an effective methods in brain disease diagnosis and detection EEG examine plays a more and more great part in clinical diagnosis.There are more and more researches in the extraction and analysis of EEG features. According to the literatures, many classic or modern signal processing methods have been applied in EEG which already have good results. Obviously it has paved a path for future research in this field. So based on the foundation there are two contents in this thesis: first, the basis theory and methods using classic and modern signal processing analysis are studied, especially the effective and fast algorithm about the extraction and analysis of EEG features; Second, the new analysis methods is developed, and combined with the existing methods applying in the extraction and analysis of EEG features. The details include:First, the theory and method about the extraction and analysis of EEG features are studied, the history and the development of various methods as well as their advantages and disadvantages are reviewed.Second, approximate entropy (ApEn) theory and its property are studied, including the development of an approximate entropy (ApEn) based analysis method for the brain injury detection and quantification. The ApEn based method is used to calculate the complexity of EEG signals obtained from the hypoxia and asphyxia experiment. The detection and analysis results show that the ApEn based method can be used as a measure to detect the injury of the central nervous system (CNS) with EEG, and the correlation coefficients between the results and the neurological deficit scores (NDS) obtained from several behavioral tests is about 90%. The ApEn based method has a potential to be a clinical measure for the injury detection of CNS.Third, the thesis analyzes epileptic EEG signals approximate entropy (ApEn) and its feasibility using in the epilepsy detection. Based on the limitations of the common used approximate entropy (ApEn) in the epilepsy detection, this thesis analyzes epileptic EEG signals with the sample entropy (SampEn) approach, a new method for signal analysis with much higher precision than that of the ApEn. Data analysis results show that the values fromboth ApEn and SampEn decrease significantly when the epilepsy is burst. Furthermore, the SampEn is more sensitive to EEG changes caused by the epilepsy, about 15%~20% higher than the results of the ApEn.Finally, this thesis proposes a new epileptic prediction method based on the experience mode decomposition (EMD) and sample entropy technologies. The data analysis results show that such techniques suppress the effects of noises and detect and even predict the outbreak of the epileptic. The new method works well and obtain a good result in epileptic prediction.