Non-invasive study of the brain
Non-invasive research is difficult, especially when you’re working with something as complicated as the brain. Imagine being at a pro sports game outside the stadium and trying to figure out what’s going on inside just by listening. I’m constantly in awe that we can record activity from the brain without breaking the skin, it’s like magic. However, it’s still difficult and not without controversy.
I do research in two labs, both of them do non-invasive research. My main-PI (university) does brain-computer interfacing primarily, while my Co-PI (hospital, where I’m about to start working full-time) focuses on spinal cord injury and neurophysiology. My work bridges the two worlds and I’m very excited to live on the border between two very different, but maybe not that different, fields. Of course, it means learning twice as much just to get half of the work done, but I still find it very satisfying.
When working my my Co-PI’s lab research is done by looking at responses in EMG, or muscle activity. This makes life fairly simple. EMG is a very “loud” electrical signal and easily recorded from the surface of the skin with a high degree of accuracy. There is a ratio, called the signal to noise ratio, which dictates a lot of the stuff that I do. EMG has high signal to noise ratio. That just means there is a lot of signal and low amounts of noise. This makes life easy because I can use the resulting data right away without any fancy noise removal techniques. It’s like having a face to face conversation with someone in a restaurant. Sure, there is background noise and other people talking, but you can clearly hear the person you are with. EMG makes my life easy because it’s very much a face to face conversation in a very (electrically) noisy world.
The brain is a different story. In fact, it’s orders of magnitude different. We record EMG in mV (millivolt) ranges, something like 0.1 to 0.5 mV on average, sometimes higher, sometimes lower, but mostly in that range. When we record from the brain we use something called EEG. If we go back to my sports stadium example, EEG would be like placing a lot of microphones around the stadium parking lot and trying to understand what’s going on inside based purely on those recordings. While EMG is in mV ranges, EEG is in uV (microvolt) ranges, which is 1000x smaller than mV. EEG recordings are so sensitive that we have to routinely filter out line noise, which is just noise from the electrical lines in the walls, lights, etc.
EEG has a horrible signal to noise ratio, which means there’s a lot of noise and not a lot of signal. However, we’ve gotten really good at working with EEG and in the past decade or two we’ve made incredible progress with using EEG to determine a person’s intention, diagnose diseases, and we can study different neurological conditions. While we can’t make out individual conversations, we can tell what groups of neurons are doing and we can triangulate where the activity in the brain is coming from (roughly) to help us understand which parts of the brain are doing what. These advances have come thanks to improved computational abilities, better understanding of how electrical signals travel through the body (volume conduction), and specialized math (or repurposed math really).
If you think about it, EEG is pretty amazing because the skull isn’t very conductive. In fact, the only thing less conductive than bone is fat. So we are recording groups of neurons firing together through all the tissues (the pia, arachnoid, and dura mater), through the cerebrospinal fluid, through the skull, over the skin, and somehow even with hair we manage to get recordings that are usable. I don’t want to say it’s witchcraft, but it’s the darkest of dark arts. If it weren’t for the huge body of literature that has proven that what we’re recording can be related to brain activity, I wouldn’t believe it.
Which funny enough was the inspiration for today’s post. My Co-PI remains highly skeptical that EEG can be used for brain-machine interfacing, even though we can get ~70-80% accuracy in predicting a person’s intent (at least basic walk, don’t walk), he’s still incredibly skeptical. I appreciate that about him and it forces me to work harder to convince him that something is “real.” It helps me because if I can convince him, then I can pretty much convince anyone. My main-PI has a better understanding of the science and math behind what we’re doing so he is less skeptical because he’s spent his career doing this stuff. My Co-PI primarily focuses on EMG recordings, which to him are as good as it gets, so he doesn’t quite believe that EEG does it claims to do.
We’re actually starting some work with a collaborator (who’s lab I may be switching to if my Co-PI does leave, more here) to do some invasive recordings to see the difference between the two and verify that what we’re seeing is “real” and not some weird artifact. And I will admit that EEG is full of artifacts, so we do have to be careful with how we interpret the data and a lot of the stuff we do is a stretch, but that’s why we need so much data. When we stretch like that you need to have the statistical power to back up what you’re finding or others will look at you skeptically.
So we’re about to do some invasive recording to compare with our non-invasive stuff and see how close the two correlate. These experiments have already been done, but it has been several decades (in some cases) that they were last performed, so it wouldn’t hurt to do again and more importantly it gives us a way to verify that our equipment specifically is working correctly, because while EEG has been shown to work, that doesn’t guarantee that our equipment is operating the way it should be.
Plus if I’m being completely honest, I’m not a big fan of the EEG equipment my Co-PI has, so it would be a good check for me as well. His equipment uses passive electrodes and I prefer active electrodes. Active electrodes amplify the signal right at the source of the recording before it travels down the wires to the computer. The line noise that we record (which again is the electrical noise from the lights, power lines in the walls, etc.) is typically captured in those wires, even if they are shielded, it’s not perfect.
Active electrodes amplify the signal right away bumping the signal to noise ratio up before more noise gets added in, so imagine if the signal recorded was 1 uV and a passive electrode would keep it at 1uV until it reached the computer, well we could have 5 uV worth of noise added to it along the way giving us a signal to noise ratio of 1/5 and that assumes perfect signal at the start (spoiler that will never be the case). If we amplify the signal at the point of the recording (the scalp electrode) if we record 1 uV of signal and amplify it 20x (as an example) we now have a 20 uV signal and if we add the 5uV worth of noise our signal to noise ratio is 20/5 which is waaaay better.
Now that’s a simple example, but it shows how active electrodes (which were invented around 1996, so not a super long time ago) really help improve EEG recordings. My Co-PI uses passive electrodes, the advantage to this is they are cheaper and some of the systems can be used in MRI, which is not the case for active electrodes. The active systems (to my knowledge) are not MRI compatible and since my Co-PI does a lot of MRI and fMRI studies (I’ll cover fMRI one of these days), he went with the passive system. Which is fine, it just makes my job harder on the processing side of things.
Overall I appreciate my Co-PI’s skepticism and I’m okay with the extra steps we’re taking to verify that everything we are doing is correct. We are currently working on a study involving [REDACTED] and it’s really cool because we hope to [REDACTED]. Kidding, as usual I can’t talk about the stuff we’re doing specifically, but eventually I hope to share. I’ve got like 4 papers in the works at the moment, so there will probably be a lot of sharing here sooner or later once they start getting published. Until then I only get to share the struggles of experimental setup and verification.
If you want to learn more about EEG I’ve written several posts on the topic. This post covers EEG basics and how we can record from the brain to begin with. I mentioned in this post source localization, which is a fancy way of saying triangulating where the “people” or neurons are talking and I even cover that in more detail here. That talks about ICA or independent component analysis pretty in-depth, but if you want more, this post covers ICA some and it talks specifically about dipole fitting, which is how we identify with a fair degree of certainty where the IC is located in the brain. This is because an IC is an “independent component” so you can think of it as a group of people all cheering in the same section of the stadium, we can isolate that cheering specifically and determine where in the stadium, or in this case the brain, the signal is coming from. You even get to see what I see when I’m working with the data, so it’s pretty cool if you’re into that sort of stuff.
Well that’s my overly long musings for the day on the brain and how wild it is that we can actually record information from it. I’m excited to be publishing all my research (basically at the same time now, lol) so I can share it here with everyone. I really enjoy the work I do and I hope once I get to share some of it, my regular readers will get to see why I love it so much. Who knows, maybe I’ll inspire a few people to go into the same type of research! Oh and one of the really cool things about doing non-invasive research? You get to be a participant in experiments if you want, I’ve been part of a LOT of experiments and it’s always fun to get to participate, in my opinion anyway!