The Application Research Of Complexity And Time-frequency Analysis On The Detection And Seizure Prediction Of Epileptic EEG Signals

Epilepsy is a common neurological disorder that affects approximately fifty million people according to the World Health Organization.EEG is widely used in the clinical examination of patients with epilepsy,professional physicians need manual analysis by experience.This method is obviously tedious and time-consuming,so a timely and accurate detection system of epilepsy is essential.It is extremely necessary to study automatic,efficient and accurate methods for epilepsy detection and seizure prediction by using computer.In this paper,the complexity analysis of entropy and time domain,frequency domain analysis are applied to the signal detection and prediction of epileptic seizure by using Freiburg and CHB-MIT databases.The main research contents and results are as follows:（1）The analysis of parameter optimization and comparison of signal detection capabilities for four kinds of permutation entropyPermutation entropy had some advantages including the robustness in the presence of observational,dynamical noise and low computational complexity.And it was used as the complexity feature in this paper.There are several kinds of permutation entropy in current study.In this paper,the process of obtaining optimal parameters of algorithm was discussed in detail,and then the ability of detecting the epilepsy signals for four kinds of permutation entropy was compared by using statistical test and classification.The results showed that reasonable parameter values had a certain influence on the accurate feature extraction of permutation entropy;They could effectively reflect the dynamic changes of brain activity and detect the inter-ictal and pre-ictal signals.Meanwhile,the ability of detection was equal to each other.（2）Using permutation entropy as features to research the epileptic seizure prediction and comparing the seizure prediction abilities of permutation entropyThe study proposes two early warning mechanisms based on statistical method（One-step FP and Two-step FP）,and compares the differences in predictive performance between the two,and the predictive power of four types of entropy.The results showed that this method could predict seizures before their occurrence and the average prediction horizon was 61.93 min.The average sensitivity was 94%and the false prediction rates was 0.111 h^-1.In addition,Two-step FP can significantly reduce the false prediction rates compared with One-step Fp,and the permutation entropy based on Shannon entropy was more suitable for early warning than the other three kinds of permutation entropy.（3）The potential of signal detection based on time domain and frequency domain signalsTwo binary classification problems and one three-class problem were designed to compare the difference between the time domain signals and the frequency domain signals by using the convolutional neural network.The results showed that using frequency domain signals in the Freiburg database,average accuracies of 96.7,95.4,and 94.3%were obtained for the three experiments,while the average accuracies for detection in the CHB-MIT database were 95.6,97.5,and 93%in the three experiments.Using time domain signals in the Freiburg database,the average accuracies were 91.1,83.8,and 85.1%in the three experiments,while the signal detection accuracies in the CHB-MIT database were only 59.5,62.3,and 47.9%in the three experiments.Experiments showed that the detection based on frequency domain signals were significantly better than the time domain signals.（4）The research of epileptic seizure prediction based on the convolutional neural networkThe deep learning algorithm based on convolutional neural network showed good results in epilepsy classification.In this study,convolutional neural network combined with Two-step FP was used to research seizure prediction.Although some patients received satisfactory results,the overall predictive performance was poor.