Written by Travis M. Moore
Last edited 16-Jun-2020
While correctly placing electrodes will certainly allow you to record voltage at the scalp, the signal we wish to record is buried in the electrical noise of all ongoing brain activity. So how do we get to the squiggly lines we see on the computer screen? Actually being able to see meaningful information requires a series of steps all its own. While up until this point we have focused on the difficulty of the voltage reaching the scalp, now we focus on the difficulty of emphasizing that tiny voltage and reducing the vastly stronger electrical noise. There are four main techniques that are commonly used in all evoked potential systems:
It should come as no surprise that the scalp voltage needs to be amplified in order for us to use it, but the problem of amplification is more complex than simply increasing the voltage. Think back to psychoacoustics when we discussed the concept of signal and noise. There we talked about the signal-to-noise ratio, or SNR. An example of this concept is trying to understand one talker in a mix of different talkers. The talker of interest is the signal, while the others can be considered noise. If we simply increase the level of the signal does the talker of interest become clearer? No, because we are turning up the other talkers as well. This is one reason why hearing aids provide limited benefit in noise. Similar to multiple talkers, there are multiple groups of neurons constantly firing in the brain. Those neurons are important and necessary, but when they get in the way of our signal of interest, they become unwanted noise in the recording. Differential amplifiers provides one method of emphasizing the signal.
If you think about any medical test that uses electrodes from T.V., do you just see a single electrode? No - there's always a complicated looking mess of electrodes and wires. While we don't always need dozens of electrodes, we do need at least three to record anything meaningful. If you think back to our discussion of electricity and voltage (module not online yet), we learned that voltage is a relative measure: voltage is always in comparison to some reference. Even in this module when we discussed the resting potential of a neuron, the inside of the neuron was -70 mV compared to the fluid outside the cell. If we compared the inside of the neuron to a different reference, the number would no longer be -70 mV.
To understand what a differential amplifier does, think back to the acoustics module about additive synthesis. Recall that when two sine waves are added together there is constructive and destructive interference. Essentially, if two sine waves are in phase, they will create a sine wave twice as large when added. If two sine waves are completely out of phase (i.e., by 180 degrees) then their sum will cancel out to zero. The same principles apply to electromagnetic waves, and we can take advantage of this behavior when recording those waves at the scalp. Differential amplification just flips the input of one of the electrodes 180 degrees out of phase to cancel out unwanted noise.
The terminology here can get confusing, mainly because there are several names for these two electrodes. The names that make the most sense regarding the differential amplifier are noninverting and inverting electrodes. You can think of the noninverting electrode as the "target" electrode, and the inverting electrode as the "reference" electrode we'll be comparing the target against. The noninverting electrode is placed to get the strongest signal from how the dipole is oriented, and the inverting electrode is placed so that it picks up as little of the dipole as possible.
Let's bring everything together to see the genius of the differential amplifier.
Keep in mind using a differential amplifier doesn't solve all of our problems. Why? Well, the noise isn't exactly the same at the two electrodes, and there is still some voltage from the inverting electrode that will interfere with the noninverting signal. But fear not, we have two more tricks up our sleeves.
Once again you'll have to think back to the module on acoustics, specifically the discussion of filters. (It's almost like there was a legitimate reason to learn all that acoustics stuff...) Years of research has shown that the potentials we measure at the scalp are relatively slow (i.e., low frequency). Because we know this, one way to remove a huge portion of the unwanted electrical noise is to filter it out. There are very specific filter settings for each potential that get rid of the most noise possible without filtering out any frequencies in the signal of interest. Because you already know (or can review) how filters work, there's not much more to say about this technique here. You'll come across the specific filter settings as you learn about each test.
One of the reasons noise is so difficult to get rid of is because it's unpredictable. If we knew exactly what everyone in a group of people was going to say at the exact time and level, it would pretty easy to make a hearing aid that could counter those talkers. It's the randomness that thwarts our attempts to control and eliminate noise. However, we can also use that randomness to our advantage, and that's where signal averaging comes in.
The noise in our case consists of random voltages, which can be either positive or negative. Do you know what the mean of -1 and 1 is? It's 0. That's right, more cancellation. Let's look at this on a larger scale. The following example was carried out using R, a programming language for statistics. Using a random number generator, let's get 1,000 values between -1 and 1, then find the mean.
x = runif(1,000, -1, 1)
x = runif(100,000, -1, 1)
The point of artifact rejection for electrophysiology is the same as for otoacoustic emissions (OAEs; online module not yet available). An "artifact" refers to something in the data that doesn't belong. Say for example your patient coughs during testing and there is a huge spike in the voltage that buries the signal of interest. Not good. Needless to say, artifacts need to be removed from the data whenever possible. Artifact rejection refers to settings in the software that will automatically exclude sweeps that exceed given criteria. We know that huge voltage spikes couldn't possibly be neural in origin, so we can tell the software to kick out any sweeps above a certain voltage.