A practical guide to the selection of independent components of the electroencephalogram for artifact correction.
Chaumon M., Bishop DVM., Busch NA.
BACKGROUND: Electroencephalographic data are easily contaminated by signals of non-neural origin. Independent component analysis (ICA) can help correct EEG data for such artifacts. Artifact independent components (ICs) can be identified by experts via visual inspection. But artifact features are sometimes ambiguous or difficult to notice, and even experts may disagree about how to categorise a particular component. It is therefore important to inform users on artifact properties, and give them the opportunity to intervene. NEW METHOD: Here we first describe artifacts captured by ICA. We review current methods to automatically select artifactual components for rejection, and introduce the SASICA software, implementing several novel selection algorithms as well as two previously described automated methods (ADJUST, Mognon et al. Psychophysiology 2011;48(2):229; and FASTER, Nolan et al. J Neurosci Methods 2010;48(1):152). RESULTS: We evaluate these algorithms by comparing selections suggested by SASICA and other methods to manual rejections by experts. The results show that these methods can inform observers to improve rejections. However, no automated method can accurately isolate artifacts without supervision. The comprehensive and interactive plots produced by SASICA therefore constitute a helpful guide for human users for making final decisions. CONCLUSIONS: Rejecting ICs before EEG data analysis unavoidably requires some level of supervision. SASICA offers observers detailed information to guide selection of artifact ICs. Because it uses quantitative parameters and thresholds, it improves objectivity and reproducibility in reporting pre-processing procedures. SASICA is also a didactic tool that allows users to quickly understand what signal features captured by ICs make them likely to reflect artifacts.