Automated identification of epileptic seizure EEG signals using empirical wavelet transform based Hilbert marginal spectrum

Abstract

This paper proposes a new method for the classification of epileptic seizure electroencephalogram (EEG) signals. Empirical wavelet transform (EWT) based Hilbert marginal spectrum (HMS) has been derived. In order to segment the Fourier spectrum of the EEG signals, the scale-space representation based boundary detection method has been employed. Then, EWT is used to decompose EEG signals into narrow sub-band signals and HMS of these sub-band signals have been computed. For a synthetically generated multi-component frequency modulated signal, the EWT based HMS is compared with the conventional Fourier spectrum obtained using fast Fourier transform (FFT) algorithm. Three features have been extracted from these HMSs which belong to distinct oscillatory levels of the EEG signals and probability (p) value based feature ranking is performed. Finally, the selected features are fed to random forest classifier for classifying EEG signals of seizure and seizure-free classes. We have achieved 99.3% classification accuracy with only 50% training rate which shows the usefulness of the proposed method for the classification of epileptic seizure EEG signals.