Epileptic State Classification based on Intrinsic Mode Function and Wavelet Packet Decomposition

The scalp electroencephalogram (EEG) signal based epileptic seizure analysis has been comprehensively studied in the past. But existing researches are generally concerned with the seizure/non-seizure detection. Few attention has been paid to the epileptic preictal state classification, which is found to be evidently more important to the injury prevention. In this paper, we study the epileptic preictal state classification for seizure prediction. The one hour preictal EEG signal is divided into non-overlapped equilong segments. Statistical features of the first intrinsic mode function (FIMF) of the empirical mode decomposition (EMD) of the EEG signal as well as the 4-level wavelet packet decomposition (WPD) of the FIMF are extracted for the EEG signal representation. A five-state classification problem is formulated, including one interictal, three preictal, and one seizure states. Experiments on the benchmark epilepsy EEG database collected by the Children’s Hospital Boston and the Massachusetts Institute of Technology (CHB-MIT) using several popular classifiers are provided for the effectiveness demonstration.