Research On The High-Frequency Oscillations Using Stereo-Electroencephalograms From Patients With Temporal Lobe Epilepsy

Epilepsy is a neurological disease caused by sudden abnormal discharge of brain neurons,leading to transient brain dysfunction.About 30% of patients with epilepsy are refractory epilepsy that cannot be controlled by medication.For refractory epilepsy,it can currently be treated by surgically removing the epileptogenic zone.High-frequency oscillations(HFOs)in EEG signals are a kind of biomarkers for localizing epileptogenic zone.At present,there is no recognized high-frequency oscillation detection method,and the subjective judgment of the doctor is still the most commonly used at present.However,human visual recognition is not only extremely subjective,but also time-consuming and labor-intensive,a new method for detecting high-frequency oscillations is required.Studies have shown that physiological activities also produce high-frequency oscillations,while only pathological high-frequency oscillations are associated with seizures.Therefore,distinguishing the physiology and pathology of high-frequency oscillations is an important research focus at present.This thesis takes stereotactic electroencephalogram(SEEG)data of patients with temporal lobe epilepsy as the object and studies a new high-frequency oscillation detection method.On this basis,through the analysis and research of high-frequency oscillation signals,the rhythmicity index and the phase coherence help distinguish between physiological and pathological high-frequency oscillations.Finally,the multi-dimensional characteristics of high-frequency oscillation signals and machine learning methods are used to further analyze it,and the relationship between the above characteristics and the effect of epilepsy surgery is established.Specific work includes the following:First,the SEEG data of patients with temporal lobe epilepsy is pre-processed,which is used for bipolar processing to remove the interference of baseline drift and other interferences of the EEG signals.The ensemble empirical mode decomposition(EEMD)method is used to obtain the intrinsic model function with high frequency band.The decomposed intrinsic model function is analyzed by the high-frequency oscillation detection method developed in this study to extract the occurrence time and average frequency of high-frequency oscillations,frequency standard deviation,duration,average amplitude,amplitude standard deviation,rhythmicity index,and phase coherence.Then,this study analyzed the correlation between the two new features of high frequency oscillations’ rhythmicity index and phase coherence and the surgical treatment effect of patients.The results show that the high-frequency oscillations’ rhythmicity index in the surgical resection area is greater than the high-frequency oscillations’ rhythmicity index in the non-surgical resection area;phase coherence can effectively eliminate harmonic interference.The support vector machine(SVM)was used to classify the long-term high-frequency oscillations in the temporal lobe surgical resection area and the non-surgical resection area high-frequency oscillation.The area under the curve(AUC)of the receiver operating characteristic curve(ROC)reached 0.96.When the characteristics of the rhythmicity index were excluded,the AUC value dropped to 0.69.The results show that the rhythmicity index is the factor that has the greatest impact on SVM classification.The rhythmicity index can be used to distinguish physiological high-frequency oscillations from pathological high-frequency oscillations.Finally,this study further analyzed the duration characteristics of the detected high-frequency oscillation data set and located 34 long-term high-frequency oscillations with a duration of more than 1 second.32 high-frequency oscillations located in the first 1/3 time periods,and there are 33 long-term high-frequency oscillations which located in the temporal lobe area.The study shows that long-term high-frequency oscillations at the early stage of temporal lobe epilepsy are very likely to be biomarkers to distinguish pathological high-frequency oscillations from physiological high-frequency oscillations,and have important clinical value.