Research On Automatic Detection Of Epileptic EEG

Epilepsy is one of the most common diseases that cause troubles to human beings,the morbidity rate of which is fairly high. From the angle of electrophysiology, thepathogenesis of epilepsy could be explained as an enormous and hypersynchronousactivity of the nerve cells. Tonic-clonic seizure and unconsciousness could be inducedby those activities. Electroencephalography (EEG) is one of the most effectivetechniques for diagnosing epilepsy. Particular wave forms generated by epilepticseizures, such as spikes, sharp waves and spike-and-sloe wave, could be reflected inEEG.Abundant information of ictal and interictal EEG patterns, which can be acquiredby long-term monitoring, is usually required for diagnosis of epilepsy. Thisinformation helps physicians to decide whether drug therapy or surgical therapy isneeded. As the amount of long-term EEG is very enormous, visual inspection of it isvery tedious and time-consuming. Besides, the process of manual diagnosis might bevery subjective. Therefore, automatic detection of epilepsy is of great importance.Two different kinds of automatic detection methods based on nonlinear featureextraction are proposed and the effectiveness of them is verified by experiments. First,some effective methods of Time-frequency analysis and nonlinear are introduced inthis paper. Then the methods above are put forward:1. Automatic detection methodbased on sample entropy and extreme learning machine.2. Automatic detection methodbased on detrended fluctuation analysis and extreme learning machine. In the firstmethod, wavelet transform is used in the first place to extract EEG of useful frequencydomain, then sample entropy of the extracted EEG signals is calculated as the featurevector, and then the feature vector is fed into ELM for classification. In the secondmethod, scaling exponents of EEG signals are calculated by detrended fluctuationanalysis. And fluctuation index of EEG signals extracted by wavelet transform iscalculated as a linear feature. These two kinds of features are combined and then fed toELM. The EEG data used in this study comes from Department of Epileptology ofBonn University, Germany. Results show that the proposed methods achieve not only high detection accuracy but also fast computation speed.