EEG recordings are always relative recordings. By “relative” this means that the recording electrode determines brain wave activity in comparison to a reference electrode. It the reference electrode is on the ear lobe, then the brain wave activity seen will be compared with the activity in the ear lobe. Surprising as it may sound, occasionally, a fair bit of brain activity can get into the ear lobe, cancelling out “real” brain activity while injecting fake activity into other recording channels. Real is in quotes because there technically isn’t actually real brain activity that is measured as it is always the perceived activity as measured against some sort of reference. There has been criticism of a referential linked-ears montage for processing EEG. So, many clinicians are using Laplacian or whole-head average montages. But in certain situations, these metrics are very misleading and will lead to a complete misdiagnosis, given the way they generate their “reference.” Please see the example below.
Below are several montages showing that when there are several channels of fairly phase-locked activity, both Laplacian and whole-head averaging can really skew the results. See below. Whenever in doubt, the clinician should be zooming in to the raw and use the ruler function to see just how phase-locked the different channels appear to be.
I had a case recently where high delta appeared along the vertex. It is my thought that a “vibrating” sterno-mastoid or neck muscle was throwing in the fake delta via the reference. I’m still waiting for muscle verification from the clinician on that. There are often neck issues and muscle spasms in those with concussion and a muscle spasm would throw fake delta into the EEG. Given the clinician is expecting to see delta as coincident with a TBI, that clinician will succumb to the streetlight effect. The streetlight effect is where a clinician or researcher only sees what is under the light – or what they expect to see. In this case, the clinician will notice the obvious delta and assume that it is real.
Referring to the example picture below:
Referential: When there are several channels of hyper-coherent activity as is often seen with alpha, we have to consider some things. First, are the waveforms phase-locked? To see this may require zooming into just a few seconds across the entire computer screen. Then use the ruler and look at the peaks and troughs for alignment. If there is some mis-alignment, then I assume the activity is real. The other thing to watch for is; if there is a maximal phase-locked rhythm along the midline (FZ, CZ & PZ), out of phase with posterior channels (in this example - alpha) then some of the real alpha signal from the posterior channels leaked in through the reference. Given volume conduction, the sites furthest from the reference are least similar to the reference and therefore show the most fake EEG. This is a primarily important assessment to make on a referential montage. Other montages can make up fake frontal alpha – in this example – as a result of vector addition (especially full-head average as seen below). In this example using a referential montage, there is next to no alpha showing up in the frontal channels, including the midline. So, I will interpret that the EEG activity is true and not caused from a corrupted reference electrode.
Laplacian: Laplacian looks at the difference of the site of interest and the difference between it and the average of the electrodes surrounding it. So, in this case, the high alpha channels look similar to themselves and therefore cancel out and show a flat line for activity. The frontal channels show activity because both the beta activity and the EMG are not phase or frequency-similar and therefore stand out against their neighboring channels.
Whole-head Average: This montage basically takes the average of all the channels and makes it the “reference” electrode through software. This too, distorts the real activity and generates fake EEG activity as well. In this case, the back 7-channels (parietal, temporal and occipital channels) are fairly similar. So let us assume they were 10 uv roughly on average. But these channels are getting averaged into the front channels which are fairly flat-lined in alpha. This will pull the average down to 7 out of 19 total channels. The alpha will then average out to 7/19 = 3.6 uv. Therefore, O1 and O2 which are higher than all other channels show alpha activity. Any channel with ~3.6 uv on it, will show as a flat line for alpha. C3, CZ and C4 are fairly close to the average and therefore flatline in alpha. Notice that T3 and T4, which are fairly flat in the referential montage, appear to have more activity in this averaged montage because they have less than the average. If they were flat-lined in the referential montage, they would show up as 3.6 uv and out of phase (red line) with all the other alpha activity. In reality, this simply is not true. Notice that the frontal channels (which are fairly flat-lined in referential) now show a few microvolts out-of-phase with the posterior channels. This is a complete error! Given these channels are fairly flat with respect to alpha, vector addition will apply to them the amplitude of the whole-head average alpha and put them out-of-phase with the posterior alpha.
Vertex: The issues with a vertex montage is obvious. Everything is subtracted against the vertex. If the vertex had alpha on it similar to the other back channels, then that alpha would be subtracted out and the back channels would show close to 0 uv. In the referential montage, CZ appears to have some random-ish beta and possibly some EMG on it. Of course, this activity spills into all the other channels as it has been subtracted out of CZ, (which the software has made “0”). Any channel that was in phase with CZ would show a flat line where the similarities coincide. Any channel with out of phase activity to CZ, would now show as double its height as seen in the referential montage.
So Laplacian and Average montages are great for highlighting an aberrant channel. These montages can be misleading when there are several hyper-coherent channels in the recording.
