Research On Focal EEG Signal Recognition Based On Deep Learning And Transfer Learning

Epilepsy can cause great harm to the patient as a common neurological disease.For example losing consciousness,causing movement disorders,physiological disorders.According to the statistics of the world health organization,there are about 50 million people suffering from epilepsy.Drug-resistant focal epilepsy patients account for about one-third of the total number of patients.Surgical treatment is the best choice for patients with drug-resistant focal epilepsy to achieve complete seizure freedom.The location of epileptogenic area plays an important role in preoperative evaluation.However,the traditional methods that artificial analysis of electroencephalogram not only time-consuming but also susceptible to personal subjective consciousness.As a result,the epileptogenic area cannot be correctly located.Therefore,it is necessary to search for an efficient and accurate technology for automatic recognition of EEG.Combining deep learning and transfer learning for focal EEG recognition was proposesed in this theisi.Firstly,the time-frequency analysis was be used in the original EEG.Then,transfer learning the AlexNet and the GoogleNet model.The last three layers of the AlexNet model were replaced with the full connection layer that contains two nodes,Softmax layer and classification layer.The output of the seventh full connectivity layer of the transfer learning AlexNet model is taken as the features of the EEG.The last three layers of the GoogleNet model were replaced with the NewDropout layer,the full connection layer that contains two nodes,Softmax layer and classification layer.The output of the NewDropout layerof the transfer learning GoogleNet model is taken as the features of the EEG.Finally,the extracted features are sent into the SVM,LSTM,BP,SRC LDA and the Softmax classifier of the network.The best classifier was selected.The focal EEG recognition algorithm which combines deep learning and transfer learning was tested and evaluated in two public EEG databases in this theisi,.When the AlexNet and GoogleNet was transfered in Bern-Barcelona database,.Comparing the result of the six classifier.The focal EEG recognition by SVM is better than other classifier.When the AlexNet and GoogleNet was transfered in the short database of UBonn.The focal EEG recognition by transfering the AlexNet model is better than transfering GoogleNet model.At the same time,the algorithm in this theisi is compared with several methods with locating epileptogenic area,and it is proved that the method in this theisi can improve the recognition effect of the focal EEG recognition.