A Study Of Functional Connectivity And Causal Network Characteristics Of Multi-channel EEG Of TLE Via Frequency Domain Granger Causality Analysis

Objective:This paper analyzes the excessive discharge of epilepsy, by functional connectivity of multi-channel EEGs, which corresponding to interictal, preictal and ictal periods were obtained for patients with temporal lobe epilepsy, and were obtained for normal control group.Methods:1patient data:All patients underwent presurgical monitoring in the Epilepsy Monitoring Unit at TianJin Medical University General Hospital, temporal lobe epilepsy group (aged19to47) by clinical diagnosis and normal control group (aged24to30) are8cases respectively. Epileptic foci of each patient was located in the anterior temporal area (scalp electrodes F7and F8). All16-channel EEGs were collected from four groups, that are the interictal, the preictal and the ictal groups, and normal control group, each group with10pieces of EEGs, that length is10seconds.2Data preprocessing:Remove the baseline drift and the50Hz frequency interference, and remove obvious artifacts by applying FastICA algorithm.3Time-Frequency Analysis:EEGs corresponding to ictal, interictal, preictal periods and normal control group were analyzed by short-time Fourier transform. Get frequency range of energy concentration in ictal periods:δ and θ.4Application bandpass filtering to obtain δ and θ of EEG frequency components.5δ and θ of EEG frequency components are respectively Granger causality analysis(1)16-channel EEGs time series were analyzed by frequency-domain Granger causality, and γij was calculated on each node.(2) Calculate betweenness centrality on each node in the causality network.(3) K-means clustering analysis:BC of temporal lobe epilepsy group were analyzed by K-means clustering. Depending on the value, each node was divided into active and inactive node category, analysis the change tendency of the active node BC from interictal period to preictal period to ictal periods.(4) Analysis causal density of network:calculate the causal density value of ictal periods. And BC and CD value of ictal periods were analyzed by k-means cluster.(5) Analysis causal flow of network:Computing causal flow value of active or inactive node in ictal periods, and causal flow value corresponding to the node in interictal and preictal periods.Results:1Time-frequency analysis:Energy of EEG during temporal lobe epilepsy ictal period, is concentrated in the low frequency:δ and θ bands, and each channel has energy distribution.2Multi-channel EEG frequency domain functional connections Delta spectrum:temporal lobe epilepsy group:DTF mean value of interictal (7.3404±1.9629); DTF mean value of preictal (4.8755±1.0541); DTF mean value of ictal (8.1770±1.6978); DTF mean value of Normal control group (2.1591±0.5561)。 Delta spectrum:temporal lobe epilepsy group:DTF mean value of interictal (8.1263±0.4356); DTF mean value of preictal (3.5955±0.8756); DTF mean value of ictal(8.2826±0.4899); DTF mean value of Normal control group(2.7373±0.6542).3BC corresponding to ictal, interictal, preictal periods were analyzed by K-means clustering.In active nodes, BC in ictal period of delta frequency (0.0499±0.0149). BC in interictal period of delta frequency (0.0469±0.0095). BC in preictal period of delta frequency (0.0800±0.0200).In active nodes, BC in ictal period of theta frequency(0.0467±0.0108). BC in interictal period of theta frequency (0.0437±0.0076). BC in preictal period of theta frequency (0.0651±0.0126).4BC and the CD value in ictal period were analyed by k-means clusteringAccording to the BC in ictal period k-means clustering analysis, each node was classified into two categories:active node1and non active node1. According to the CD value in ictal period k-means clustering analysis, each node was classified into two categories:active node2and non active node2. Each frequency band in the delta and theta, active node1and2active node coincidence rate is very high, both are the same, or active node1includes the active node2, or2includes active active node1, Active node1, or2, are not consistent with epileptic foci, active nodes1, or2, rarely contain epileptic foci. Active node in the ictal period was analyed causal flow.5Cause flow analysis of active nodeDelta frequency:CF values of active nodes in ictal period (0.6864±0.3037), CF values of active node corresponding to the interictal,ictal period of regional (0.1495±0.1358)、(0.1174±0.0648)Theta frequency:CF values of active nodes in ictal period (0.6661±0.1645), CF values of active node corresponding to the interictal,ictal period of regional (0.0774±0.1973)、(0.1489±0.0570)Conclusions:1The advantage frequency band of EEG in ictal period of temporal lobe epilepsy was low frequency band.2The function connection feature of temporal lobe epilepsy group enhancements, corresponding to the normal control group. And the feature in ictal period than interictal, preictal period, is stronger.3BC of active nodes in ictal period of temporal lobe epilepsy is obviously enhanced, than in interictal, preictal period. And BC are no significant difference between in the interictal and preictal period.4Active node, are obtained according to the causal density and BC K-means clustering analysis, are similar.5Causal flow of active nodes in ictal period of TLE is larger, the nodes are the causal source. Causal flow of inactive nodes in ictal period of TLE is smaller, the nodes are the causal sinks.