Research On Recognition Algorithm Of Epileptogenic Foci EEG Signals Based On Complex Domain Analysis

Epilepsy is a disease of the cranial nervous system characterized by persistent epileptic tendency.In the world,about 15 million epilepsy patients are drug-resistant,and require surgical removal of the epileptic area to control or cure epilepsy.In this way,accurate positioning of the epileptogenic focus is the key to the successful surgical treatment.EEG can directly reflect the electrophysiological activity of the brain,which is a necessary means for the preoperative diagnosis of epileptic areas.The localization of these areas is usually conducted by clinicians with the visual inspection of 24-hour long-term electroencephalogram(EEG)on patients.However,such means is time-consuming,and labor-intensive with subjective and empirical characteristics,which has made the epileptogenic area auxiliary diagnosis technology based on digital signal processing become the hotspot in the epilepsy research field currently.The most critical task of this technology is to design an epileptogenic foci EEG signal recognition algorithm based on signal processing and pattern recognition according to the characteristics of EEG during epileptic activity.On the one hand,it overcomes the weaknesses of manual diagnosis and locates the seizure-onset area accurately and efficiently,and alleviates the pain of patients.On the other hand,it also lays the foundation for the development of the follow-up epilepsy auxiliary diagnosis system.However,the complexity and diversity of epileptic EEG bring challenge to achieve accurate and efficient recognition of the epileptogenic foci EEG signals.Aiming at solving the problem of poor accuracy,efficiency and generalization ability in the existing recognition algorithms of epileptogenic foci EEG signals,this paper designs the recognition algorithms based on the complex domain analysis according to the non-stationary and nonlinear characteristics of epileptic EEG signals,attempting to explore the effectiveness of the proposed algorithm on the auxiliary localization of the epileptic foci.The main research work and innovative findings of this paper are as follows:(1)Aiming at the insufficient characterization capabilities of the real number domain on EEG phase information,the paper proposes an epileptogenic foci EEG recognition model based on flexible analytic wavelet transform.The EEG signal in real domain is decomposed into a series of complex EEG sub-band coefficients by flexible analytic wavelet transform,and the amplitude and phase information of EEG signals are retained simultaneously.The complex-valued distributed entropy is introduced to realize the synchronous mining of the amplitude-phase nonlinear information of the complex-valued EEG coefficients.The proposed algorithm model combines the flexible time-frequency representation of flexible analytic wavelet transform,the nonlinear analysis ability of complex-valued distribution entropy and log energy entropy to mine the potential pathological information effectively and more fully,which enhances the recognition performance of the proposed model.In differentiating 3750 pairs of focal and non-focal EEG signals over Bern-Barcelona database,the proposed algorithm has achieved 95.26% recognition accuracy,96.35%specificity and 94.21% sensitivity,verifying the effectiveness of the complex domain analysis method in the identification of EEG signals in epileptic area preliminarily.(2)Aiming at the poor mining adaptability of the traditional time-frequency analysis method on EEG rhythm information,the paper proposes an epileptogenic foci EEG recognition model based on the complex-valued EEG rhythm features.This model combines the dual-tree complex wavelet transform and the Hilbert transform to obtain the amplitude and phase modulation information of the EEG rhythm in the complex domain;subsequently,it combines the standard deviation,singular value and the complex-valued fuzzy distribution entropy to capture the difference information of the complex EEG rhythm from multiple perspectives so as to deeply reveal the rhythm characteristics of the focal and non-focal EEG signals.Finally,the Logit Boost algorithm is applied to integrate the decision tree classifiers to enhance the reliability and stability of the classification results.The experimental results reveal that the δ+θrhythms have the greatest contribution to the recognition of epileptogenic area as well as the largest feature difference.For 3750 pairs of focal and non-focal EEG signals,the proposed algorithm obtained 98.83% recognition accuracy,0.976 Matthew correlation coefficient,and 8.1ms single-sample recognition time.It indicates that the proposed algorithm based on complex domain analysis can mine EEG rhythm information more fully,and achieve the balance between recognition performance and computational complexity,which improves the accuracy and efficiency of the auxiliary localization in the epileptogenic area.It further shows the superiority of EEG recognition algorithm based on complex domain analysis.(3)Aiming at the poor stability and generalization of the feature learning of the traditional EEG feature extraction algorithm,the paper proposes an epileptogenic foci EEG recognition model based on the amplitude-phase fusion matrix and the deep feature learning network.On the basis of the above research conclusions,the 0-8Hz EEG frequency band is obtained by low-pass filtering of EEG signals directly.Meanwhile,Hilbert transform is performed on the filtered EEG signal,and the amplitude matrix and phase matrix of the analytical signal are obtained.The Multi-scale guided filter fusion is introduced to fuse the amplitude matrix and phase matrix in data layer to obtain the amplitude phase fusion matrix containing deep information.Furthermore,PCA neural network is utilized to automatically learn and adaptively select features directly from the amplitude-phase fusion matrix,which solves the difficulty of feature design caused by individual differences and overcomes the experience and deviation of manual feature design.Finally,two EEG data sets of Berne-Barcelona and Bonn are employed to verify the proposed algorithm.The experiment results on Bern-Barcelona and Bonn EEG databases reveal as follows: the recognition accuracy of EEG signals in epileptic areas is 100%;the Matthews correlation coefficient is 1;for 7 different epilepsy detection tasks,a recognition accuracy of more than 99% and a Matthews correlation coefficient of 0.975 can be obtained.The experimental results verifies that the proposed method ensures the recognition accuracy and computational efficiency,and shows better generalization ability and robustness in different epilepsy diagnosis task scenarios.In summary,this paper takes focal and non-focal EEG signals as the research object,and conducts research on the construction of the most critical epileptogenic foci EEG signal recognition model in the auxiliary positioning technology of EEG signals based on EEG signals.Through applying the complex domain representation model of EEG signals,the accurate and efficient recognition of the epileptogenic areas is achieved,which plays a positive and effective role in promoting the practical process of the auxiliary localization technology in epileptogenic area,and lays a theoretical foundation for the development of the epilepsy’s auxiliary diagnosis intelligent system in the next step.