Computerized pattern recognition of EEG artifact

Publisher Website
Google Scholar
Abstract
Summary Automated artifact classification of quantified EEG (QEEG) epochs from 9 males using linear discriminant analysis showed greater than 85% agreement with judges' opinions. These results were replicated (n=600 epochs for each sample). Testing the entire sample (n=5800) illustrated reliable eye artifact (94%) but reduced muscle artifact classification (70%) accuracy. Agreement was lowest in the case of more subtle forms of muscle artifact (i.e., low amplitude muscle), however, less than 4% of these were wrongly classified as non-artifact. Improved data collection techniques retaining high frequency energies are anticipated to improve muscle artifact recognition. Results indicate that low levels of artifact contamination would result when only those epochs classified as non-artifact were accepted for inclusion in further analysis.