Abstract: High frequency oscillations have been recently proposed as a new biomarker of seizure onset zone. Marking epileptic discharge is highly time-consuming, easy to get false detection, and the relation between discharge and seizure still leaves much to define. This paper presents a fast automated detection method based on the difference of power spectral density in each channel, and similarity of filtered signal enveloped on epilepsy propagation path, and also the combination with complex Morlet wavelet transform to classify high frequency oscillations.The results obtained in 4 clinical cases show high performance in terms of sensitivity and specificity which can be considered as an effective method in preoperative localization.

Key words: high frequency oscillations； epileptic recordings； AR model； complex Morlet wavelet transform