| Epilepsy is a common neural system disease,it is normally characterized by transient brain dysfunction caused by sudden discharge of brain cells.EEG contains a lot of physiological and pathological information,clinically doctors diagnose the illness through visual inspection of EEG for 24 hours.The necessity of EEG in epilepsy diagnosis and the deficiencies of visual detection make automatic detection technology based on signal processing the hotspot of research currently.The core task of computer automatic detection technology is to realize the analysis and interpretation of EEG signals through signal processing and pattern recognition methods.Online seizure diagnosis system has been developed abroad and applied clinically successfully,while in China,researches still remain at the laboratory stage.The main reason is that most current epileptic EEG analysis algorithms are based on single domain with low level of excavation and representation of effective pathological information,making it difficult to meet clinical requirements with analytical effect,and there are still problems with the stability,adaptability and generalization of algorithm.Under this circumstance,it is particularly urgent and necessary to seek an efficient and dependable epileptic EEG signal analysis method.The article starts with the characteristics of the epileptic EEG signals,researches the differences of EEG signals during different stages of attack,and proposes analysis algorithms in complex domains on this basis.Regarding the issues in terms of the stability,adaptability and generalization of existing testing algorithms,in-depth research is carried out to set up corresponding algorithm models based on composite domain analysis and explore the application of the proposed algorithm to the automatic diagnosis of epilepsy.The main work and innovation of this article are as follows:(1)Regarding the unstable result of classification and great data fluctuation caused by changes in epilepsy discharge,the article proposes an epileptic EEG analysis model based on wavelet-envelope spectrum.The model is based on double-density discrete wavelet transformation,combines envelope analysis algorithm,makes use of the complementary advantages of algorithm to disclose more characteristics and information and improve the specificity of information.On the sound basis of complex domain,in order to relieve the contradictory relationshipbetween the complexity and recognition rate of algorithms,several simple statistical parameters are extracted to describe the differences between the interictal and ictal EEG signals,while ensuring the representation ability of characteristics,the complexity of algorithm is also reduced,avoiding the heavy burden of characteristic calculation in single domain algorithm.In addition,at the recognition stage,Adaboost algorithm is made use of to integrate neural network,further improving the feature learning ability and results of classifiers.The algorithm organically combines the algorithms in time-frequency domain and frequency domain,and has a preliminary exploration of the application of complex algorithm domain to processing EEG signals,and the result of the experiment shows that the accuracy of the algorithm in testing the EEG signals during the interictal and ictal is 98.93%,while compared with the method of single algorithm domain analysis,the fluctuation of the result is smaller and the recognition rate is more stable.(2)Regarding the poor adaptability in the switch between different classifications of tasks by the algorithm,the article proposes an epileptic EEG analysis model based on multi-order and fractional order Fourier transformation-wavelet packet decomposition(Fr FT-WPT).The algorithm unites the two time-frequency analysis algorithms of Fr FT and WPT to calculate the fuzzy entropy of various sub-bands for the quantitative analysis of the non-linearity of EEG signals in complex time-freuqnecy algorithm domain.In order to make fuller use of the information after fractional Fourier transformation decomposition,the characteristics of adjustable number of orders of its parameter,multi-stage fractional Fourier transformation idea is proposed,through the adjustment of focus level of Fr FT,different information of various focus levels of epilepsy EEG is captured.Considering the different combinations of the effective characteristics of different classification tasks,on the basis of principal component analysis and Kruskal-Wallis test,an improved characteristic selection plan is proposed,for self-adaptive optimal representation of different classification tasks.The algorithm recombines the algorithms with two time-frequency domains and tests seven common classification tasks,repeated experiments show that the accuracy of the classification result of the algorithm for seven tasks is all more than 99.0%,switching between different tasks without loss of performance,proving good adaptability.(3)On the basis of the methods during the early stage,the article further studies the ability of recognizing and generalizing the data of different patients by complex domain analysis,and proposes a novel EEG analysis model based on MODWT-KDE,and makes use of a great deal of clinical data of many patients to verify the performance of the algorithm.In the framework of this algorithm,first of all,lead choice is made with the method based on minimum standard deviation and maximum mutual information,and intercepts EEG data by sliding of non-overlapping rectangular windows;meanwhile,kernel density estimation algorithm is introducedfor the statistical analysis of MODWT decomposition coefficient,beneficial for the expression of statistical characteristics while the complementary characteristics of two algorithms also weakens the influence of wavelet functions on classification performance.During the classification stage,synthetic minority oversampling algorithm is adopted to avoid the phenomenon of over-fitting due to the over concentration and singularity of learning information,so as to improve the generalization ability of classifiers.The algorithm has combined the information of the space domain,time-frequency domain and statistical domain,shows outstanding robustness in verification,and the average sensitivity and average specificity of 23 subjects are respectively 97.84% and 99.97%,perfectly stable results can still be maintained in the recognition of a great deal of data of many patients,proving excellent generalization application potential.Above all,the paper takes epileptic EEG as the subject of research,and discusses the algorithm design issue of EEG analysis as the core of epilepsy automatic detection technology.Through the research of the fluctuation of data in single recognition of algorithm in fixed classification task,the switch and adaptation of the algorithm between different tasks of classification and the recognition and generalization of the algorithm for different patients,the methods of analyzing epileptic EEG in composite domain is discussed.Regarding the above-mentioned three questions,composite domain analysis is utilized to achieve accurate and stable epilepsy detection,the work of the paper can promote the development and application of the automatic diagnosing technology of seizure,providing new thinking and effective solutions for the development of computer-aided diagnosis system of diseases of the brain. |
PDF Full Text Request