Research On Automatic Epilepsy Detection Based On Bidirectional LSTM Deep Network

Epilepsy is a chronic,repetitive brain disorder and a nervous system disease that is extremely difficult to cure.The quality of life of epileptic patients and their families is severely reduced by the impact of repeated seizures due to the irregular and excessive discharge of neurons.Electroencephalography(EEG)is used as an important guiding tool in clinical diagnosis or treatment.EEG contains abundant physiological and pathological information.At present,it is a commonly used clinical method to diagnose epilepsy by manually interpreting EEG of patients with epilepsy.Manual diagnosis is time-consuming,tedious and mixed with the subjective judgment factors of doctors.Therefore,the development of an automatic epilepsy detection system and automatic analysis of EEG can reduce the burden of doctors and improve work efficiency,which is of great significance for the diagnosis and treatment of epilepsy.This paper presents a method of automatic epilepsy detection based on bidirectional LSTM(Bi-LSTM)deep network.Firstly,the local mean decomposition technique with good performance for analyzing non-stationary signals was used to decompose the original EEG signals into a series of product functions with physical significance,and the first three product function components were selected for statistical calculation.Then,they are fed into the constructed Bi-LSTM deep network model for feature learning and classification recognition.The Bi-LSTM deep network can be modeled in two directions,suitable for analyzing long-term EEG signals.At the same time,the training parameters of Bi-LSTM deep network are constantly adjusted to achieve the optimal state of EEG signal analysis;Finally,the output of the deep network is optimized by post-processing technology,which can further improve the accuracy of automatic detection.The seizure detection method proposed in this paper achieved an average sensitivity of 93.61%,an average specificity of 91.85%,and an average G-average of 92.66% on the EEG data of 24 patients with 877.39 hours.At the same time,compared with the LSTM network,the Bi-LSTM deep network constructed in this paper has significant advantages in the comparison experiment.Compared with other published results,we can find the excellent performance of the proposed method in the automatic detection of seizures.In addition,the method proposed in this paper needs to be tested on more reliable EEG data in order to fully verify the feasibility of its clinical application.