the recent emergence and success of electroencephalography (EEG) in low-cost portable devices, has opened the door to a new generation of applications processing a small number of EEG channels for health monitoring and brain-computer interfacing. These recordings are, however, contaminated by many sources of noise degrading the signals of interest, thus compromising the interpretation of the underlying brain state. In this paper, we propose a new data-driven algorithm to effectively remove ocular and muscular artifacts from single-channel EEG: the surrogate-based artifact removal (SuBAR).

by means of the time-frequency analysis of surrogate data, our approach is able to identify and filter automatically ocular and muscular artifacts embedded in single-channel EEG.

in a comparative study using artificially contaminated EEG signals, the efficacy of the algorithm in terms of noise removal and signal distortion was superior to other traditionally-employed single-channel EEG denoizing techniques: wavelet thresholding and the canonical correlation analysis combined with an advanced version of the empirical mode decomposition. Even in the presence of mild and severe artifacts, our artifact removal method provides a relative error 4 to 5 times lower than traditional techniques.

in view of these results, the SuBAR method is a promising solution for mobile environments, such as ambulatory healthcare systems, sleep stage scoring, or anesthesia monitoring, where very few EEG channels or even a single channel is available.