Automatic artifacts removal from epileptic EEG using a hybrid algorithm

Electroencephalogram (EEG) examination plays a very important role in the diagnosis of disorders related to epilepsy in clinic. However, epileptic EEG is often contaminated with lots of artifacts such as electrocardiogram (ECG), electromyogram (EMG) and electrooculogram (EOG). These artifacts confuse EEG interpretation, while rejecting EEG segments containing artifacts probably results in a substantial data loss and it is very timeconsuming. The purpose of this study is to develop a novel algorithm for removing artifacts from epileptic EEG automatically. The collected multi-channel EEG data are decomposed into statistically independent components with Independent Component Analysis (ICA). Then temporal and spectral features of each independent component, including Hurst exponent, skewness, kurtosis, largest Lyapunov exponent and frequency-band energy extracted with wavelet packet decomposition, are calculated to quantify the characteristics of different artifact components. These features are imported into trained support vector machine to determine whether the independent components represent EEG activity or artifactual signals. Finally artifact-free EEGs are obtained by reconstructing the signal with artifact-free components. The method is evaluated with EEG recordings acquired from 15 epilepsy patients. Compared with previous work, the proposed method can remove artifacts such as baseline drift, ECG, EMG, EOG, and power frequency interference automatically and efficiently, while retaining important features for epilepsy diagnosis such as interictal spikes and ictal segments.