Research On Seizure Prediction Based On Deep Learning And Scalp EEG

Epilepsy is a common chronic central nervous system disorder with random recurrent involuntary seizures that severely affect the lives of patients and their families.Traditionally,the diagnosis of epilepsy has relied on medical practitioners to test and analyse the EEG of patients with epilepsy.However,due to the complexity and diversity of EEG waveforms,the time consuming nature of the tests and the heavy reliance on the expertise of health care professionals,as well as the limited human effort,the diagnosis is prone to errors,and therefore automated epilepsy prediction is of great research importance.It has been found that the electroencephalography(EEG)of epileptic patients produces different waveforms at different stages of the body,and therefore the EEG signal can be divided into different periods,which provides a theoretical basis for epileptic seizure prediction.In this paper,we build a deep learning model to automatically extract epileptic EEG data features and classify the signals of different epilepsy periods,so that we can build a unique seizure prediction scheme for different patients.In this paper,the EEG signals of each epilepsy patient in the Children’s Hospital Boston epilepsy scalp EEG dataset(CHB-MIT)were first filtered,segmented and extracted from the preictal and interictal periods for labelling,and transformed into two-dimensional EEG signals with temporal and frequency characteristics by Short-time Fourier transform(STFT).The STFT is used as input to the Convolutional Neural Networks(CNN)for subsequent epilepsy prediction model training and prediction.Secondly,personalised prediction models for epilepsy patients were developed for the CHBMIT dataset.Based on the characteristics of different epilepsy patients’ data,a suitable epilepsy prediction model was designed.In order to alleviate the problem of model overfitting due to the small amount of data,this paper fine-tunes and improves the model on the original pre-processing model based on the migration learning idea,so as to speed up the training process of the model.In this paper,five migration learning pre-training models(VGG16,Alex Net,Goog Le Net,Resnet18 and Efficient＿b0)were used to extract the EEG signal features of epilepsy patients.The seizure prediction model based on migration learning and SVM was implemented.Finally,the paper uses the Vision Transformer(Vi T)to build a personalised epilepsy prediction study model for each patient on the CHB-MIT dataset.The image features are then extracted by the Transformer Encoder module,and finally the MLP Head module is used to personalise the seizure prediction.This paper investigates the impact of different deep learning models on epilepsy prediction studies based on epileptic EEG signals and deep learning,and evaluates the performance of different models in predicting epilepsy,providing a new idea for future seizure prediction studies in epilepsy.