Research On Seizure Detection Method Based On Transfer Learning And Gated Recurrent Unit

Epilepsy is one of common nervous system diseases.Repeated and sudden seizures have a great impact on patients’ life,and even cause serious harm to their physical and mental health.In recent years,the technology research on epilepsy diagnosis and treatment has attracted much attention.In order to reduce the time cost of manually interpreting EEG in epilepsy diagnosis and to improve the detection efficiency,automatic seizure detection methods based on transfer learning theory and gated recurrent unit neural network have been studied in this paper.The main research works are as follows.（1）An automatic seizure detection method is proposed based on bidirectional gated recurrent unit（Bi-GRU）network.After performing a time-frequency decomposition and feature extraction on long-term multi-channel EEG,a Bi-GRU neural network is established,with its outputs post-processed such as moving mean filtering,threshold judgment and seizure fusion.The Bi-GRU neural network can efficiently capture the long-term dependence and correlation in EEG sequences,and made the proposed seizure detection method obtain a good detection performance in about 867 hours of testing data from CHB-MIT scalp EEG dataset.The average sensitivity and recognition rate are 93.89% and 98.49%respectively.（2）To solve the problems such as unbalanced EEG data and difficulty in training deep network,a seizure detection method based on transfer learning of VGGNet-16 is proposed.The VGGNet-16 network pre-trained on image dataset is transferred and combined with a gated recurrent unit network to realize the pattern recognition of epileptic EEG.On CHBMIT EEG dataset,this seizure detection method obtained an average sensitivity of 90.12%and an average recognition rate of 96.29%.The seizure detection methods proposed in this paper have good detection performance and clinical application prospects,and the research based on gated recurrent unit and transfer learning can provide a new idea for the development of seizure detection technology.