Research On Epilepsy Classification Based On Dynamic Brain Function Network

Epilepsy is a recurrent neurological disorder induced by multiple factors.During seizures,brain neurons are repeatedly over-discharge,resulting in intermittent dysfunction of the central nervous system.The unpredictability of seizures greatly affects the lives of patients with epilepsy and even endangers their lives.Therefore,the study of epilepsy has important clinical significance.EEG is widely used in cognitive neuroscience and psychological research,which not only contains a lot of information about brain function,but also reflects the cortical dysfunction of many diseases.The main means of EEG analysis is the combination of EEG and neuropathology.Therefore,it is often used to explore the mechanism of brain function and to describe the relationship between EEG rhythm and neuropathology.The complete state transition of epileptic stages is not a static process,but the change of brain synchronization caused by the change of stages is an important sign.these dynamic functional changes will lead to changes in local connectivity of the brain network.How to identify the dynamic changes of epilepsy before the onset of seizures has practical significance.In this dissertation,a method of epilepsy classification based on dynamic brain functional network is proposed.The dynamic brain functional network of epilepsy in different frequency bands is constructed based on EEG signals,and their staging differences and synchronization changes are discussed.Furthermore,the network properties are calculated,and the staging differences of properties in different frequency bands are described by nonparametric tests.Finally,the interictal and preictal are classified based on support vector machine,and a good classification effect is obtained.To provide strong support for the next step of epilepsy early warning.The main research work of this dissertation is as follows:(1)The sliding time window method and weighted phase lag index are introduced to construct dynamic brain functional networkIn this dissertation,the EEG data of 17 patients from CHB-MIT epilepsy dataset of Boston Children’s Hospital were selected.First,the data were preprocessed,and then the sliding time window method was used to divide the 6s window without overlap.Based on the EEG signals of delta,theta,alpha,beta and gamma bands,18 EEG channels were used as network nodes,and the synchronization between nodes is calculated by weighted phase lag index which can reduce the influence of small phase difference and irrelevant noise interference.The dynamic brain functional network is constructed in interictal,preictal,ictal and postictal stages.The difference and synchronization change of functional connection during pairwise comparison were analyzed by two-sample t-test.The results show that the connection differences and synchronization changes are mainly reflected in the alpha,beta,and gamma bands,and become more significant as the frequency band increases.The results provide a new perspective for the study of the dynamic changes of epilepsy.(2)To analyze the differences of dynamic brain functional network properties in pairwise comparisonAfter constructing the dynamic brain functional network,the clustering coefficient,characteristic path length,global efficiency,local efficiency and small-world property of different stages in alpha,beta and gamma were calculated based on graph theory.And the network property differences of interictal,preictal,ictal and postictal stages were analyzed by nonparametric tests.The results show that the characteristic path length and global efficiency are significantly different in most frequency bands,while the clustering coefficient,local efficiency and small-world attributes are significantly different only in some frequency bands.It is proved that the network properties can reflect the changes of epileptic stages to a certain extent.(3)Classification of interictal and preictal based on support vector machinesThe feasibility of the method proposed in this dissertation is verified on CHB-MIT dataset.In the experiment,a personalized classification model based on support vector machine was constructed for each patient.The results of accuracy,specificity and sensitivity analysis showed that whether it was dynamic functional connection index or network property index,it can achieve good results in interictal and preictal classification,especially in high frequency band.For specific patients,up to 100% accuracy,100% specificity,and 100%sensitivity can be achieved.For all patients,the dynamic functional connection feature can achieve an average accuracy of 90.9%,specificity of 92.6%,sensitivity of 89.2%,and the optimal network property feature(clustering coefficient)can reach an average accuracy of91.0%,specificity of 91.2%,and sensitivity of 90.9%.This provides important support for epileptic prediction and more accurate identification of preictal state in the future.