Automated detection of electroencephalography artifacts in human, rodent and canine subjects using machine learning

• Published in: Journal of neuroscience methods Vol. 307; pp. 53 - 59
• Main Authors: Levitt, Joshua, Nitenson, Adam, Koyama, Suguru, Heijmans, Lonne, Curry, James, Ross, Jason T., Kamerling, Steven, Saab, Carl Y.
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 01.09.2018
• Subjects: Artifact detection; Electroencephalography; Machine learning; SVM; Electroencephalography; Artifact detection; SVM; Machine learning
• ISSN: 0165-0270; 1872-678X; 1872-678X
• DOI: 10.1016/j.jneumeth.2018.06.014

•EEG artifacts, if not properly and objectively detected, can confound statistical analysis of quantitative EEG.•We developed a fully-automated method for detecting EEG artifacts in humans, rodents, and canines using a Support Vector Machine (SVM) approach.•The method detects several artifact features across species with high accuracy. Electroencephalography (EEG) invariably contains extra-cranial artifacts that are commonly dealt with based on qualitative and subjective criteria. Failure to account for EEG artifacts compromises data interpretation. We have developed a quantitative and automated support vector machine (SVM)-based algorithm to accurately classify artifactual EEG epochs in awake rodent, canine and humans subjects. An embodiment of this method also enables the determination of ‘eyes open/closed’ states in human subjects. The levels of SVM accuracy for artifact classification in humans, Sprague Dawley rats and beagle dogs were 94.17%, 83.68%, and 85.37%, respectively, whereas 'eyes open/closed' states in humans were labeled with 88.60% accuracy. Each of these results was significantly higher than chance. Other existing methods, like those dependent on Independent Component Analysis, have not been tested in non-human subjects, and require full EEG montages, instead of only single channels, as this method does. We conclude that our EEG artifact detection algorithm provides a valid and practical solution to a common problem in the quantitative analysis and assessment of EEG in pre-clinical research settings across evolutionary spectra.