The detection of artefacts in EEG series.

In a clinical trial of an anticonvulsant drug, series of electroencephalogram readings are sometimes obtained. These may contain artefacts, that is patches of atypical values which must be identified and either removed or compensated for before a full analysis can be carried out. Methods for identifying such artefacts are discussed. These include non-parametric methods and also parametric ones combining the ideas of autoregressive modelling and of influence in regression. These methods are described and illustrated on typical data sets.