A Study On EEG-based Automatic Epilepsy Detection

Epilepsy is a chronic neurological disorder of the brain that affects around 50 million people worldwide.Electroencephalogram(EEG)is a useful tool for epileptic seizure detection.There are many characteristic waves in EEG during epileptic seizures,which is currently the main basis for the diagnosis of epileptic disorders.At present,the clinical diagnosis of epileptic disorders are mainly achieved by visual inspection of neurological experts.It is time-consuming,inefficient and lack of standards.Automatic detection of epileptic EEG signals can solve the problem.In this paper,the following three aspects of research work have been carried out on the EEG-based automatic epilepsy detection:1)A detection method of epileptic characteristic waves based on constructed wavelet function is proposed.In this method,wavelet functions are constructed from epileptic characteristic waves in patient's EEG signal,and continuous wavelet transform is applied to EEG signal in appropriate scale range.Next we normalized the transform coefficients,which represents the similarity between constructed wavelet function and original EEG signal.Then we used single threshold method detect the occurrence of epileptic characteristic waves at that position.The experimental results show that the detection performance using constructed wavelet function is better than the common wavelet functions which are widely applied to the detection of epileptic seizures.2)A detection method of epileptic characteristic waves based on multi-wavelet synthesis is proposed,which is a further improvement on the method based on single constructed wavelet.Because of the significant differences among epileptic characteristic waves,the method based on single constructed wavelet has some limitations.In this thesis,we tried to synthesize the information of several wavelet functions and developed the “transform before synthesis” method and “synthesis before transform” method.We compared these two method with the method based on single constructed wavelet function.The experimental results show that the “transform before synthesis” method improve the detection performance effectively,and the sensitivity of the detection of epileptic characteristic waves is higher than 90% and the erroneous detection being lower than 2.5%.3)An identification method of seizure region based on wavelet and relative amplitude feature is proposed.We didn't make full use of the information in transform coefficient when the single threshold method was applied.Thus in this thesis,we propose a SVM classification algorithm based on the distribution features of the transform coefficients and the features of relative amplitude in EEG signals.Finally,the seizure regions are identified,which has been applied to 10 patients including 34 seizures.An accuracy up to 94% has been achieved.