Automatic recognition of random artifacts in recordings of slow brain potentials: a bias-free method

A method for computerized recognition of random artifacts in recordings of slow brain potentials is presented. It is based on systematic comparison of the shape of single trial EEG segments with all other trials obtained in the same experiment. This is possible because, in contrast to short-latency event related potentials, a plateau of EEG negativity during complex tasks can often be detected in single trails. After linear trend removal and subtraction of EOG artifacts, similarity of pairs of single trials is established by point-to-point linear regression analysis, the total score of a trial being defined as the mean correlation coefficient (R) resulting from comparison to all other trials. Selection is then performed by setting a critical value for R, admitting for averaging only trials with a mean correlation coefficient above the critical value. An example is given in which the method is applied to recordings of slow brain potentials during a manual tracking task.