For better or worse, I wrote a traditional dissertation instead of a more practical “staple” dissertation composed primarily of journal articles. Hence there are some potentially useful things in it that will likely never be read beyond my committee and, perhaps, people willing to search the thing piecemeal on Google Books. Perhaps this blog can improve the situation a bit.
Anyway, for my thesis I had the fortune of working with Scott Makeig, who pioneered the use of ICA for EEG analysis (Makeig, Bell, Jung, & Sejnowski, 1996). ICA is the best general-purpose tool out there for correcting for EEG artifacts such as blinks and muscle activity. Artifact correction in general isn’t perfect (Groppe, Makeig, & Kutas, 2008), but in practice ICA artifact correction works quite well (Mognon, Jovicich, Bruzzone, & Buiatti, 2011).
One issue facing ICA newbies though is learning to recognize which independent components (ICs) represent artifacts and which reflect true neural activity. In case it helps here are few prototypical examples of artifact ICS. In each figure is the IC’s scalp topography, activity ERPimage, and power spectrum. If you load an EEGLAB dataset into EEGLAB, you can make a plot like this for each IC using Plot->Component properties in the EEGLAB GUI menu (see below).
How to make plots of IC properties like those below using the EEGLAB GUI.
An IC corresponding to blink activity. The topography and large monophasic humps in the ERPimage are the most characteristic features of this artifact. Usually these ICs in EEGLAB have low numbers (like IC 1) because their magnitude at the scalp is so large (EEGLAB sorts ICs from biggest to smallest scalp magnitude–more or less).
This IC corresponds to horizontal eye movements. Note that the polarity of the IC flips across the eyes and the ERPimage typically has square wave-like jumps. These ICs typically have low numbers as well.
An IC corresponding to muscle activity (EMG). These ICs typically have a lot of power in the 20-50 Hz range and focal topographies that project to only a few neighboring electrodes. Sometimes they flip polarity across neighboring electrodes.
If a mastoid reference electrode is near an artery, the artery will move the electrode producing monophasic, brief blips that occur about once a second. This IC captures that artifact as apparent by the blips in the ERPimage and the topography with a strong right mastoid weight.
Makeig, S., Bell, A. J., Jung, T.-P., & Sejnowski, T. J. (1996). Independent component analysis of electroencephalographic data Advances in Neural Information Processing System, 145-151
Groppe, D. M., Makeig, S., & Kutas, M. (2008). Independent component analysis of event-related potentials. Cognitive Science Online, 6 (1), 1-44
Mognon A, Jovicich J, Bruzzone L, & Buiatti M (2011). ADJUST: An automatic EEG artifact detector based on the joint use of spatial and temporal features. Psychophysiology PMID: 20636297