Research On Feature Extraction And Automatic Detection Of Epileptic EEG

EEG signals involve a great deal of information about the function of the brain, which relate to the different physiological states of the brain. And thus it is a significant for understanding the complex dynamical behavior of the brain in clinical application, Brain and Cognitive sciences. Epilepsy is the most common neurological disease and it may be causes by many pathological processes of genetic or acquired origin. Epileptic seizures may lead to transicent disturbances of mental function and movements of different body parts. The devastating features of epilepsy are patients suffer much pain and damage for their body and mind, so much as endanger their lives. Especially for kids with epilepsy, development of their body and mental being stunted, thereby result in society problems. EEG is an important tool for the detection of epilepsy by analysis abnormal activity of EEG recordings. Traditionally EEGs are scanned by experienced physicians. This process is very time consuming, especially in the case of long term EEG recordings. Disagreement among readers of the same record is common due to the subjective nature of the analysis. Therefore, automatic techniques have become necessary for prompt and accurate identification of epileptic activities.This research has made systemical analysis for feature extraction and automatic detection of epileptic EEG. It refers to sampling EEG signals, removing artifacts and detection methods of milti-resolution and nonlinear deterministic dynamics. The approach of classifying detection is proposed and a higher detection rate is reached by the experiment performance. Besides, Support Vector Machines are applied to classify different types of epileptic EEG and the conclusion is drawn that nonlinear classifier of SVM has got a high recognition rate. Finally, the paper focuses on the study of EEG signal compression, real-time processing and feature extraction. The result demonstrates that the method is effective for EEG data compression and feature extraction. The propresed method for dada processing, feature extraction and automatic detection is helpful to realize an automatic and portable detection system for epileptic EEG.The main topics studied in this thesis are as follows:Research on the methods of removing artifacts from EEG. In automatic detection circumstances, the method of removing artifacts should be considered to provide a clear EEG signal in real time for the further processing. Extended Infomax ICA is applied to remove eye blink from clinical EEG signals by the analysis of time domain feature, topography and power spectrum. The result demonstrates that the method is effective for removing artifacts. Research on the automatic detection method based on milti-resolution. The feature extraction based on wavelet decomposition combined with sub-space energies is applied to analyze the epileptic avtivity in time domain, frequency domain and space distribution using adaptive threshold. The results demonstrate that the proposed method can detect sharp wave and spike-slow wave compared with eye movement, and is effective for feature extraction.Research on the classifying detection method based on nonlinear deterministic dynamics. Considering EEG signal is a typical non-statinary signal. On the basis of the summarization of the excellences and limitation of featrure extraction method, the thesis presents a method based on wavelet decomposition and approximate entropy to perform feature extraction, classifying detection and numerical analysis. The effectivity can be proved by the results of recognition by Neyman-Pearson criteria in clinical EEG data.Research on the algorithm of epileptic EEG classification. The classification is to yield the corresponding according to the feature extraction. The thesis presents linear classifier and non-linear classifier of support vector machine (SVM). The linear classification method is compared with Neyman-Pearson and results are consistent. Also, the results prove that higher classification ratio is reached by non-linear classifier.Research on algorithm of the multi-channel EEG processing. It is significant for data compression in automatic detection and real-time processing system. A method based on principle component analysis (PCA) and wavelet transform combined with ApEn is applied to process multi-channel EEG signal. Moreover, it is helpful to speedly give the location of epileptic activity on the electrode. Therefore, the work can provide a strong theoretic foundation of realizing a portable and desirable detection system for epileptic EEG.The research in this thesis has a certain instructional significance to study on the feature extraction and automatic detection in the epileptic EEG processin. And partial results from this work also can be extended to other signal processing and analysis. The adaptability and robustness of the researched methods need to be further verified due to the lack of more experimental data. Finally, conclusions of the dissertation are deduced and the future work is discussed.