Part 2: Mainly on eye tracking and pupillometry data.
intention
Controlling for eye-related artifacts in EEG is a standard protocol, so I wanted to compile a list of the sources of electrical activity that result in distortion of EEG signal.
we may not need to remove artifacts
I outlined in my last post a couple of examples using deep learning models to either “learn noise” in order to remove it, or utilize noise with the end goal of assisting classification. There are similarities in how researchers use the word “noise” to how words like “artifacts” or “distortions” are also used, in my opinion.
Including distortions in EEG may be essential to some research. A quick example of this is Tevis Gher's project mentioned in the last post, where correction of blinking and other eye tracking related events would remove useful information that could aid in classification.
I just mean to note that aggressive filtering or thresholding when finding independent components can lead to cases where we remove brain activity that affects the system of interest (e.g. EEG) and another that produces “artifacts” (e.g. blinking) to the detriment of the goal of a study (e.g. decision-making in a visual task where activity in certain regions of the brain affect blinking rates). In the parenthesized example of blink rate, it may be important to not remove blink artifacts when studying attention in children with ADHD, since Caplan et al 1 showed it may be a critical discriminatory measure between ADHD and non-ADHD populations.
eye related artifacts
With that out of the way, I wanted to summarize the types of corrections that seem to be most common in EEG processing, related to eye tracking data. This is based on (and the quotes below derive from) a very nice, controlled study by Plöchl, Ossandón and König in 2012 2.
| artifact | what is it | what causes it (quotes from Plöchl et al) |
|---|---|---|
| corneo-retinal dipole movement | large ocular movements | orientation change of the eyeball and thus of the corneo-retinal dipole produced between the negatively charged retina and the positively charged cornea |
| blinks (eyelid-induced) | spontaneous or willful blinking | eyelid slides down over the cornea, which is positively charged with respect to the forehead. Thereby the lid acts like a “sliding electrode,” short-circuiting the cornea to the scalp and producing artifacts in the EEG signal |
| eyelid-saccades | saccade-accompanying ballistic eyelid movements | occur in synchrony with the rotation of the eyeball and therefore are not distinguishable from the corneo-retinal dipole offset in the raw data. During upward saccades, for instance, eyelid and eyeball move upwards with approximately the same speed. However, after the termination of both, eye- and eyelid saccades, the eyelid continues to slide more slowly for another 30–300 ms and produces a signal change that is observable particularly after upward saccades |
| saccadic spike potential | potential appearing right before saccade onset | depends on the type of eye movements. Characteristically, it's a biphasic waveform, starting around 5ms prior to saccade onset and consisting ofa larger positive deflection followed by a smaller negative deflection |
| microsaccades | involuntary, small eye movements that occur during fixations | microsaccade-induced confounds in the EEG are mainly caused by spike potentials occurring at microsaccade onset, while orientation changes of the corneo-retinal dipole only play a minor role |
correction of eye artifacts
This will be covered in the next post.