Epilepsy Seizures Prediction Based On Neural Network Methods

Epilepsy is a chronic disease in which transient brain dysfunction is caused by sudden abnormal discharges of brain neurons.Epilepsy is characterized by recurrent episodes.During the onset,neurons emit abnormal EEG pulses that cause brain and body behavioral abnormalities.The unpredictability of seizures harms the physical,mental,and cognitive performance of epilepsy patients.If a more effective method can be found to predict the patient’s seizures,then the risk would be minimized by timely alert and prompt intervention before the actual attack.At present,the diagnosis of epilepsy is mostly judged by experienced medical personnel visually observing brain waves combined with some characteristic clinical manifestations.First of all,this paper conducts an in-depth investigation and summary of the research status of seizure prediction methods,and clarifies the latest research progress and future development direction.The epileptic seizure prediction strategies based on traditional method,supervised method,unsupervised method and semi-supervised method is introduced in detail.Secondly,this paper proposes a seizure prediction method based on the Generative Adversarial Networks.In order to achieve a better seizure prediction effect than traditional methods,this paper uses deep learning methods to predict seizures.Since the deep learning method relies on a large amount of training data which is difficult to acquire sometimes,this paper solves this issue by Generative Adversarial Networks architecture,which makes deep learning method can be effectively applied to seizure prediction tasks.Firstly,one-dimensional EEG signals are converted into twodimensional spectral feature images;then,EEG features are learned and generated through Generative Adversarial algorithms;finally,EEG features are sent into Extreme Learning Machine for classification.This method was performed on the intracranial EEG datasets published jointly by the University of Pennsylvania and the Mayo Clinic containing two groups of patients with epilepsy.Compared with other methods,the proposed method achieved higher classification accuracy.Finally,this paper further improves on the basis of proposed “Generative Adversarial seizure-attack prediction method”,and proposes a seizure prediction method based on Semi-Supervised Generative Adversarial Network.Semi-supervised learning methods can use labeled data and unlabeled data at the same time,reducing the degree of human involvement.Since the unsupervised method’s adaptive characteristics may lead to uncontrollable generation directions,this paper enhances the reliability of the model by adding a classification constraint to the generated model.The use of Semi-Supervised Generative Adversarial Network makes the mapping of samples with same class more compact in the feature space,effectively constraining the distribution of samples in the feature space,also achieving higher classification accuracy.