Generalized Morphological Component Analysis for EEG source separation and artifact removal

To remove artifacts from multi-channel electroencephalography (EEG) data, we propose the use of generalized morphological component analysis (GMCA). GMCA separates the EEG signals into sources that have different morphological characteristics. Each source is sparse in an overcomplete dictionary, which is constructed using discrete cosine transform, Daubechies wavelet basis and Dirac basis. The sources related to artifacts are then removed. Semi-simulated EEG signals of movement-related potentials trials contaminated by eye-blink and muscle artifacts are used to evaluate the algorithm´s performance. The performance of GMCA is compared with those of two other blind source separation algorithms, AMUSE and EFICA. The results demonstrate that GMCA successfully removes artifacts from EEG signals and the resulting distortions in both time and frequency domains are significantly lower than those of the other algorithms.