Database-driven artifact detection method for EEG systems with few channels (DAD)

We demonstrate a method for identifying and removing electroencephalograph (EEG) artifacts from mobile Brain Computer Interface (BCI) data with few channels. The main components of the method entail; one-time selection of visually inspected “good quality” EEG data from past experiments and outlier detection using statistical methods. The standardly used thresholding and filtering method was compared with the DAD method on 32 datasets of real daily-life EEG data and 660 simulated EEG datasets using sensitivity and specificity measures. Both methods detected artifacts within a similar range, however the DAD method was more accurate at deciphering artifact from brain related activity within both real and simulated datasets. Additionally, the DAD method can identify specific types of artifact signatures that directly relate to behavior. The identified artifacts/behaviors include; bitting, lead artifact, electrode pop-off, electrode artifact, electrode and lead artifact, high frequency artifact, and alpha activity.