Analysis Of Time-frequency Features Of Epileptic Intracranial Electroencephalogram Signals

Epilepsy is a common chronic neurological disorder.The underlying seizure mechanism has not been clarified.An effective method for inhibiting seizures in patients with refractory focal epilepsy is surgical resection of the epileptogenic zone(EZ).The effect of surgical treatment is closely related to the accuracy of the EZ localization.At present,intracranial electroencephalogram(EEG)signal is the gold standard for localizing the EZ.Visual inspection of intracranial EEG signals by experienced clinicians is the primary method for clinical localization of the EZ,which is subjective and time-consuming.Quantitative analysis of intracranial EEG signals may help clinicians understand the characteristic of signals and identify the EZ more quickly and objectively.Therefore,it is of great significance to quantitative analysis of intracranial EEG signals.Three time-frequency features of intracranial EEG signals recorded by stereotactic electrodes collected before surgery in 25 patients with refractory focal epilepsy were analyzed quantitatively in this study.First,we constructed the dynamic effective network based on time-variant multivariate autoregressive model and spectrum-weighted time-variant partial directed coherence,applied graph theory to calculate the network indicator in-degree,and applied the gray level co-occurrence matrix to extract features of in-degree variation over time as brain network features.We calculated the approximate entropy and line length as single-channel signal features.Next,we defined the brain areas identified by clinicians before surgery and removed during surgery as clinical-EZs,and the clinical-EZs were considered as the EZs for patients(n=23)who were seizure-free after surgery.We performed statistical analysis of the time-frequency features of signals within and outside the clinical-EZ of the patients who were seizurefree after surgery,and explored the differences in the time-frequency features of signals within and outside the clinical-EZ.Finally,we proposed a cluster analysis method combining brain network features with singlechannel signal features to localize the EZ,and evaluated the effectiveness of the method.The results showed that these three time-frequency features used in this study had significant statistical differences within and outside the clinicalEZ of seizure-free patients.The EZ localization results showed that combining these three time-frequency features can localize the EZ effectively,and the localization results were better than those using only one of them.In summary,our study showed that the results of combining brain network features and single-channel signal features to localize the EZ were better than using only one of these features.It suggested that different features characterized the EZ from different perspectives according to its own properties.Combining these complementary features provided more precise EZ localization results.