Automated Detection Of High Frequency Oscillations And Its Application On Epileptogenic Zone Identification

Epilepsy is a common neurological disorder caused by abnormal firing of neurons in the brain and affects about nine million people in China.Among them,many patients with drug-resistant epilepsy can only rely on surgical treatment to get rid of the impact of the disease on work and life,so it is very important for patients to locate epileptogenic zone.In recent years,the study found that the High Frequency Oscillations(80 ~ 500Hz)in EEG is closely associated with epileptogenic zone,the epileptic free region may also have High Frequency oscillations,but the number of HFOs inside and outside the epileptogenic zone has certain difference.Because of the huge amount of EEG data,manual vision inspection of HFOs will occupy a lot of researchers' time.Therefore,this study proposed an automatic detection method of HFOs based on convolutional neural network,and located the epileptic foci based on the characteristics of HFOs in the channel level,in order to improve the success rate of epileptic surgery.The main contributions of this psper are as follows:First,this paper designed a HFOs manual screening and calibration system.Through the input and analysis of intracranial EEG signals of five clinical patients,a database of high-frequency oscillations with nearly 15,000 samples was constructed.The database contains basic information such as the starttime and endtime of the high-frequency oscillations in the intracranial brain of five patients.Secondly,this paper proposes an automatic detection algorithm for high-frequency oscillation signals,which use the short-time energy method(STE)to initial-detect EEG signals,converts the suspected high-frequency oscillations into time-frequency graphs,and finally use the convolutional neural network(CNN)for classification.The sensitivity and false detection rate of the detector for ripple detection(80~250Hz)are 88.16% and 12.58%,respectively.The sensitivity and false detection rate for fast ripple detection(250~500Hz)are 93.37% and 8.1%,respectively.Finally,this paper presents a new method for locating epileptogenic zone based on the characteristics of high-frequency oscillations in the channel.The algorithm works base on the HFOs detection algorithm,it will the extract features of high-frequency oscillation signals detected in each channel and use PCA to reduce dimensiona.The performance of a variety of machine learning classify methods are compared.The SVM have the best performance.When use tenfold method to verify the algorithm,the the accuracy,sensitivity and specificity were 85.05%,71.25% and 92.42%,respectively.The accuracy,sensitivity and specificity of the proposed algorithm when use leave one out to verify is 79.08%,69.75% and 89.06% respectively.The above results show that the HFOs automatic detection method proposed in this paper can analyze a large number of data in a short time and has a better detection performance.The localization algorithm of epileptogenic foci designed in this paper has a certain clinical value,which is helpful to assist doctors to accurately judge the epileptic foci region and effectively treat refractory epilepsy.