A Combined Wavelet Packet-blind Source Separation Approach for Identification and Removal of Muscle Artifacts from Electroencephalogram

Electromyogram (EMG) induced electrical activity is an undesirable interference in cerebral electroencephalogram (EEG) data. We propose an efficient algorithm for automatic detection and removal of EMG artifact, while preserving most of the true cerebral activity in the EEG. First, the EEG data are decomposed into independent components (IC) using canonical correlation based blind source separation (BSS) and the components contributing to EMG are identified and filtered using wavelet packets. Finally, inverse BSS transformation is applied for reconstructing the de-noised EEG. The efficacy of proposed technique is tested by applying it to simulated and real EEG data.