Epileptic EEG Recognition Based On Wavelet Packet Energy And Entropy Using Real AdaBoost Algorithm

Epilepsy is a transient brain dysfunction caused by paradoxical discharge of the brain neurons,whose symptoms are spasm of limbs and transient loss of consciousness.Epilepsy has been one of the most common diseases of the nervous system.When it happens,it will bring great pain to patients and it is hard to be cured.It will cause bad effects on patient’s body,spirit,marriage and social status,and also bring great burden to their families.At present,the examination methods of epilepsy include imaging examination,blood examination and electroencephalogram(EEG)examination.Specially,as the main detection method of epilepsy,EEG is still playing a decisive role in the medical field.The EEGs of epileptic patients during interictal stage and ictal stage have obvious abnormal performance in amplitude and signal complexity,which has important reference value in clinical practice.At present,EEG signals are more dependent on the epilepsy experts’interpretation by their eyes,but the long-term interpretation of EEG signals will give a burden to their analyses.In addition,it is hard for different experts to give a consistent conclusion to the same segment of EEG signals because of their experience differences.Now the developments of computer technology and machine learning theory bring new ideas to the solution of this problem,and the automatic detection and recognition technology of EEGs has become a popular research direction.In order to solve the problem of automatic detection of EEG in different periods of epilepsy,a new automatic recognition method of EEG signals based on nonlinear feature and wavelet packet energy feature extraction combined with error correction coding(ECOC)Real AdaBoost algorithm is proposed in this paper.Firstly,the features of input EEG signals should be extracted.Specifically,we calculate the entropy feature of an EEG segment,and after the wavelet packet decomposition we extract the wavelet packet energy of some frequency bands.Then the feature vector is formed and the Real AdaBoost classifier based on ECOC is used to distinguish the EEG signals from different stages.The main work and innovation of this paper are as follows:(1)A feature extraction method based on time-frequency and nonlinear feature is proposed,which takes full use of various features to describe signals in time domain,frequency domain and entropy,and improves the discrimination of EEGs in different periods.(2)In order to achieve the real-time requirement,the sample entropy(SampEn)algorithm is improved aimed at the low efficiency of original algorithm of long sequence.The further optimization is on the foundation of fast approximate 4 entropy(ApEn)algorithm based on binary matrix,and the purpose is trying to avoid the repeated computation.This optimization algorithm improves the efficiency and basically meets the real-time requirement of feature extraction.(3)In order to realize the discrimination of normal,interictal and ictal periods of EEGs,the classification method based on ECOC is used in this paper.The method extends the traditional binary classifier to three-class classifier and improves the determinant conditions,which takes the full use of the confidence information of single classifier’s output as the evidence of discriminant results.This method replaces the original discriminant method based on Hamming distance and improves the recognition rate effectively.The datasets of this research are from the University of Bonn.The database has five groups of EEG signals,which contains the data of normal people with their eyes open and closed,the data collected inside and outside of the epileptic foci from patients during their interictal period and the data from patients during their ictal period.The results show that the descriptions of an EEG combined time-frequency feature with nonlinear feature can distinguish the EEGs in various periods effectively and the method has better operation speed and stability.Compared with the similar research,the method proposed in this paper has a higher recognition rate and plays a supporting role in the detection of epilepsy in clinical practice.