Epileptic Eeg Detection Method Based On Frequency Band Electrode Correlation And LSTM

Epilepsy is caused by abnormal synchronous discharge of brain neurons.It is a common nervous system disease with sudden,repetitive and transient clinical characteristics.EEG can reflect the internal state of the brain,using EEG to predict epileptic seizures is a common means of epilepsy detection.In addition,with the development of computer technology,automatic detection of epileptic seizures through deep learning has become a hot topic in the field of epilepsy clinical research.Long short term memory（LSTM）is a kind of time recurrent neural network.It is proposed to solve the long-term dependence problem of recurrent neural network（RNN）.Problems related to time series are often dealt with by LSTM.Therefore,LSTM network is introduced to detect epilepsy.The main work of this thesis is as follows:1.Detection of epilepsy based on coherence feature.The transmission of brain signals is realized through the interaction between brain functions.The coherence of EEG signals expresses the synchronization and desynchronization of electrode pairs in different brain regions at a specific frequency in a certain period of time.Firstly,the frequency bands are divided into theta,alpha,beta and gamma bands,and then the electrode coherence of each band is calculated,and the LSTM network is used for classification.The experimental results show that the classification accuracy of the gamma band is the highest.2.Detection of epilepsy based on LBP and correlation features.LBP（local binary pattern）can obtain the local texture features of EEG signals.In this experiment,the LBP features of EEG signals are extracted by using one-dimensional local binary pattern,then the EEG signals are decomposed and reconstructed by singular spectrum analysis（SSA）,and the EEG signals are divided into the above four bands,and then the electrode correlation matrix of each band is calculated,Finally,the electrode correlation matrix is sent into LSTM network for training.The experimental results show that the classification accuracy in beta band is the highest,the accuracy is88.0509%,and the sensitivity is 89.3920%.