| After the United States,European Union,and Japan,China proposed the “China Brain Project” in 2016,and early diagnosis of neurological diseases is an important research direction of the China Brain Project.Epilepsy is a neurological disease caused by abnormal discharge of brain nerves.The frequency of seizures is high and seriously affects the quality of life and safety of patients.Among them,30% are patients with drug-resistant epilepsy,and the effective cure for these patients is surgical removal of epilepsy lesions.The key to successful surgery is how to identify epilepsy lesions and normal functional areas in the preoperative evaluation stage.Among many preoperative evaluation methods,i EEG(intracranial electroencephalography)has become the gold standard for clinical diagnosis of epilepsy because it can accurately capture the rapid state of brain activity.In this thesis,we study the brain connectivity algorithms in both frequency and time domains,and then analyze the i EEG of the relevant regions during the seizures recorded by these algorithms to obtain the effective connectivity.The connectivity network then uses group analysis and centrality analysis to perform statistical analysis to distinguish the lesion area from the normal brain area,thus providing a strong basis for the precise positioning of epilepsy lesions during surgery.NPDC(Nonlinear Partial Directed Coherence)algorithm can detect linear and nonlinear causality between signals in the frequency domain.It models data based on an autoregressive model,and usually uses FROLS(Forward Regression Orthogonal Least Squares)algorithm to estimate the model parameters.However,the FROLS algorithm has poor noise immunity and performs poorly in high-dimensional data analysis.In order to overcome this shortcoming,this thesis improves the FROLS algorithm based on Kalman filter,which is called FROKF(Forward Regression Orthogonal Kalman Filter)algorithm.The experimental results show that the FROKF algorithm shows good noise resistance and accuracy on the multidimensional linear and nonlinear model simulation data,and the NPDC algorithm which uses the FROKF method to estimate its model coefficients also robustly detects the linear and nonlinear causality between signals.Granger Causality(GC)algorithm is a classic and widely used algorithm in analyzing brain connectivity in the time domain.The robustness of the GC algorithm depends heavily on the effective estimation of autoregressive model parameters(such as model order,that is,time delay between signals,etc.).In response to this problem,in order to effectively capture the effect of the long delay between signals on causality,a neural network GRU(Gated Recurrent Unit)model that is better in the field of data modeling is used to model the data,replacing the autoregressive model in the classic GC algorithm,called GRU-GC(Gated Recurrent Unitbased Granger Causality)algorithm to detect brain connectivity in the time domain.Autoregressive model simulation experiments and physiological model simulation experiments both verify the effectiveness and robustness of the GRU-GC algorithm in detecting the connectivity of linear and nonlinear brain effective connectivity.Finally,the GRU-GC algorithm is used to analyze the effective connectivity of the i EEG collected from the suspected lesion area of the real epilepsy patient,and then the effective connectivity network of the brain region related to the seizure is obtained.After that,a group analysis method was used to classify the 12 channels(7 are O groups and 5 are P groups)of the i EEG signal,and the centrality analysis method was used to analyze the importance of each channel.The experimental results show that there are only one or two channels in the epileptic seizure stage(ictal)that are inconsistent with the recommendations given by clinical experts,indicating that the algorithm can be used as an effective method for preoperative evaluation of epilepsy,and provides a strong basis for clinicians to accurately locate epilepsy lesions. |
PDF Full Text Request