Automatic artefact detection in a self-paced brain-computer interface system

An algorithm that detects various types of artefacts in a self-paced brain-computer interface is proposed. This method achieves similar performance to the state-of-art methods but has the following advantages, 1) being fully automatic (the threshold values are automatically found), 2) does not use additional EOG, EMG or frontal/temporal EEG channels, and 3) computationally inexpensive. The data were collected from the motor cortex areas using 15 EEG signals. The features extracted include the maximum amplitude of EEG signals and the stationary wavelet transform coefficients. To detect the artefacts, a simple threshold-based classifier is applied. The experimental results demonstrate that when detecting ocular and electrode movement artefacts, the method has a sensitivity (correctly detecting segments with artefacts) of 77.7%, and specificity (correctly detecting artefact-free segments) of 82.8%. For artefacts with more pronounced effects in the high frequency bands (e.g. facial muscle artefacts), the sensitivity and specificity are 83.5% and 70.3% respectively.