A system for automatic artifact removal in ictal scalp electroencephalograms /

Scalp electroencephalograms (EEGs) constitute a well-established modality in the diagnosis of epilepsy. EEGs are frequently contaminated by artifacts originating from various sources such as scalp muscles, ocular activity, or patient movement. Recently, independent component analysis (ICA) has been applied to separate and remove statistically independent artifactual sources from scalp EEG recorded during seizures. However, this method requires a trained electroencephalographer to visually identify the artifacts among the components extracted by ICA.Proposed is a system to automate this process, using a Bayesian framework to classify the components as either brain activity or artifact. The system identified EEG components with 87.6% sensitivity and 70.2% specificity. Most misclassified components were mixtures of EEG and artifactual activity. The classification error rate was comparable to the human intra-expert variability observed in EEG classification tasks. The value of system lies in its ability to remove simultaneously and automatically several types of artifacts from the EEG