A Novel Application of Empirical Mode Decomposition (EMD) to Feature Extraction of Epileptic EEG

Electroencephalogram (EEG) is the concentrated expression of physiological activity of the brain. Effective EEG feature extraction methods are key to improving different EEG recognition rates, which is a significant issue in EEG studies. A new feature extraction method is proposed in this paper based on Empirical Mode Decomposition (EMD), which can decompose nonstationary EEG into a series of Intrinsic Mode Functions (IMFs). The new method takes full advantage of characteristics of EEG to extract more effective features. The new method includes four steps: 1) EMD decomposes original EEG into a series of IMFs. 2) The correlation between each IMF and original EEG is calculated, and the energy spectrum of each IMF is calculated in θ band. 3) The IMF with maximum correlation and IMF with biggest ratio of energy are individually selected to substitute original EEG. 4) Several different features (volatility index, variability coefficient, average frequency, and variance) are extracted from IMF with maximum correction, and the phase lock feature is extracted from IMF with biggest ratio of energy in θ band. These features are composed as the feature vector of original EEGs. Support Vector Machine (SVM) is used as a classifier and the effectiveness of the new method is evaluated based on epileptic EEG. The experimental results show that the recognition rate is 94.29% and better than other EEG feature extraction methods.