EEG, not quite mind reading

Well I’m behind schedule and even though I was given an extra week (6 days now) I don’t know that I’m going to meet my deadline. It’s not me, it’s my computer, things take time to run so I’m stuck waiting for it to do its thing. Today let’s talk about what EEG is and how we make sense of it.

EEG is short for electroencephalography, a word I somehow can spell correctly even though I can barely spell. It’s a way for us to measure the electrical activity of the brain via the scalp. We do this non-invasively, so nothing penetrates the skin, the sensor literally just sits on top of the skin, hair and all.

Of course, that would make for a poor connection so we also have to add a conductive gel in between the sensor and the scalp. This gel bridges the gap so impedance (which is basically resistance) is low. Keeping the impedance low gives us a better signal to noise ratio (SNR) and the SNR for EEG is pretty bad.

Let’s imagine you want to record the conversations of people in a stadium if we got right up into the crowd and shoved a microphone in someone’s face that is about as invasive as you can get, but you would get a good SNR because the source (the person talking) is right next to the recording device (the microphone). EEG is non-invasive though so what we do is surround the stadium with microphones and try to make sense of the noise that way.

Now we wouldn’t be able to make sense of one person talking, but say a whole group started a chant. That is what we would pick up via EEG and not only can we do something with that information, because we’re recording from several different “microphones” in several different locations we can triangulate where the noise is coming from… to a certain degree of accuracy, but enough that we have a general idea.

So keeping with our analogy if we say the crowd in the stadium represents the neurons of the brain, the stadium itself the skull, and the “microphones” surrounding the stadium are the sensors, we correlate certain activity to certain events. That doesn’t mean we can read your mind though, we don’t know what your thinking or anything like that because unlike our analogy we don’t know how to speak neuron. We don’t know the language so it’s all jumbled electrical squiggles to us (see below), but we can do certain things to make sense of it.

One of the big things we do is that we can look at the different frequencies that make up the signal. Just like a crowd of people talking, EEG signals aren’t just one frequency there is a rich spectrum that make up those squiggles. For those who don’t know what that means, below is a good explanation of adding two signals with different frequencies and amplitudes together.

The top signal (blue) has an amplitude of A1 and a frequency of f1, the second (green) has an amplitude of A2 and a frequency of f2, the sum of the two signals is in the bottom and has a third amplitude and is made up of the two frequencies. Notice that the bottom signal looks a lot more complex than either of the first two “pure” signals.

Different frequencies are associated with different states in the brain (resting, thinking, sleeping, etc). We can estimate what you are doing in the broad sense using these frequencies. We can even estimate movements and we can tell when you’re about to move BEFORE you actually move (which blows my mind). There are so many different methods for making sense of EEG you could write a book about it (and plenty of people have done just that), so I can’t cover all of them or how we pick which methods we use (at least not all in one post!).

Let’s talk limitations, we can’t tell what you’re thinking. While we’re working on both invasive and noninvasive methods to decode things like speech and thoughts in the general sense, you don’t have to worry about having your private thoughts being shared publicly. At least not yet and certainly not using noninvasive methods. The accuracy of EEG or any sort of brain-machine interface (invasive or otherwise) is still variable and can be person dependant in some cases.

There’s also a lot of environmental noise we contend with in EEG. The EEG cap can move on the scalp and this will cause noise artifacts in the data. If you are running for example, that will show up and you can actually see the artifact as a function of how fast you’re running. There’s also line noise (from electrical outlets and wires in the walls). Since we’re recording very tiny electrical changes (microvolt changes!!) any electrical interference will show up, in the US it manifests itself as a large bump at a 60 Hz frequency, in Europe it’s 50 Hz.

In short, while it’s still a crude way to decode things going on in the brain EEG is a powerful tool because it’s noninvasive. We can test on all sorts of populations that we wouldn’t be able to test with using invasive methods. More importantly, we can do it fast and change the experiment quickly.

So why am I talking about EEG today? Well because I’m processing some data I collected and cleaning the data takes forever!!!! It’s annoyingly slow even for my giant computer that I spent good money on to process datasets like this. In any case, I should get back to work, this data isn’t going to make sense of itself… yet!