Performance evaluation of empirical mode decomposition for EEG artifact removal

Electroencephalography measures the sum of the post-synaptic potentials generated by many neurons having the same radial orientation with respect to the scalp. The electroencephalographic signals (EEG) are weak and often contaminated with different artifacts that have biological and external sources. Reliable pre-processing of the noisy, non-linear, and non-stationary brain activity signals is needed for successful extraction of characteristic features in motor imagery based brain-computer interface (MI-BCI). In this work, a signal processing technique, namely, empirical mode decomposition (EMD), has been proposed for processing EEG signals acquired from volunteer subjects for characterization and identification of motor imagery (MI) activities. EMD has been used for removal of artifacts like electrooculography (EOG) that strongly appears in frontal electrodes of EEG and the power line noise that is mainly produced by the fluorescent light. The performance of EMD has been compared with two extensions, ensemble empirical mode decomposition (EEMD) and multivariate empirical mode decomposition (MEMD)using signal to noise ratio(SNR). The maximum SNR values found for EMD, EEMD and MEMD are 4.30, 7.64 and 10.62 respectively for the EEG signals considered.