Efficient Rejection of Artifacts for Short-Term Few-Channel EEG Based on Fast Adaptive Multidimensional Sub-Bands Blind Source Separation

Abstract

Artifacts rejection is crucial to electroencephalogram (EEG) application. Short-term few-channel EEG (e.g., in real-time detection of stress level and motor imagery) brings new challenges for removing artifacts due to less data. Existing artifact removal methods cannot guarantee both effectiveness and efficiency for removing artifacts from short-term few-channel EEG recordings. Consequently, we propose a fast adaptive multidimensional sub-bands blind source separation method to remove artifacts from short-term few-channel EEG recordings effectively and efficiently. Firstly, noise-assisted fast multivariate empirical mode decomposition (NA-FMEMD), as a fast adaptive multidimensional sub-bands decomposition method, is employed to decompose short-term few-channel EEG recordings into multidimensional sub-bands. Then canonical correlation analysis (CCA), as a blind source separation method, is used to estimate artifacts-related and EEG-related sources. Finally, EEG-related sources are intelligently selected and reconstructed as clean EEG recordings. The results demonstrate that our method takes at least 5 times less computing time for 2-s few-channel EEG recordings than state-of-the-art methods with similar effectiveness, using the same computer and software. Therefore, our method enhances the efficiency of removing artifacts from short-term few-channel EEG recordings while ensuring effectiveness, and it is more suitable for real-time processing.