Outlier detection to identify artefacts in EEG signals

A method is presented to identify artefacts in clinical EEG recordings. The method is based upon the assumption that artefacts manifest themselves as outliers in one or more EEG-derived parameters. Parameters used are primarily derived from a 5^th order autoregressive model, fitted to each subsequent second of EEG data. The method results in a set of threshold detection rules that identify outliers in parameter space. Detection thresholds are derived from an analysis of the experimental cumulative distribution function in a training set of clinical EEG recordings, annotated by a human observer