Epileptic Seizure Prediction based on Region Correlation of EEG Signal

The existing methods of epileptic seizure prediction usually analyze the electroencephalogram (EEG) signals in the time domain, frequency domain or time-frequency domain. Although some good results have been achieved, the research and utilization of spatial information is still insufficient. Moreover, some studies extracted different features for different patients and achieved good results, but these methods are not universal and robust. Different from the previous methods, this paper propose a new feature processing method of EEG signal. All electrode signals on the scalp are considered as a whole, and fusing data from different regions to obtain spatial information. Then the correlation of first derivatives is used to obtain fluctuation information of signal caused by epilepsy, which further enlarge difference of signal in different seizures stages. In addition, we also design a post-processing strategy, which uses time-series information to rectify prediction results, so that the final result is more accurate. Finally, experimental results from the CHB-MIT dataset show effectiveness of proposed method and strategy, while the extensive result confirms that our method is superior to several state-of-the-art methods in recent years.