The Brain is Where the Heart is

Or, Initial Forays into Building a Brain-Computer Interface at Home

Earlier this week, I hooked my head up to a bunch of wires and attempted to stream my brainwaves to my laptop. To my surprise, it worked on the first try, and nothing in science ever works on the first try. I was able to run a short test to confirm that I was capturing neural signals and not just electric noise. Notably, I didn’t do this in an academic lab or an electrically shielded space surrounded by research assistants. Instead, I was in an exceptionally ordinary room in a townhouse built in the 1980s, alone, without a technician or assistant.

In today’s entry of Principia Fantastica, we’ll start out by covering some basics of Brain-Computer Interfaces, then look more into the tech that makes reading brainwaves possible. Finally, we’ll wrap up by talking a bit more about the significance of the test I carried out as the first step in building my own at-home BCI.

Click this image to view the demo video on YouTube:

Video showing EEG data collection in a regular room. Upper right plot shows cleaned time series data from two electrodes. Lower Right plot shows FFT frequency decompositions. Note the changes in the time series and the spike at the 10 Hz mark in the FFT when my eyes are closed vs open.

What is a Brain-Computer Interface and why do I want one?

A Brain-Computer Interface (BCI) is a device that lets you interface with a computer by using signals directly from the brain. Right now, I am writing this article using a traditional computer interface. Signals from my brain are arcing through my hands to press buttons on a keyboard that allow me to control inputs to the computer. A BCI, on the other hand, would allow me to control functions of the laptop by thinking about performing actions. One common example of this is motor imagery. For example, when you think about clenching your fist, a unique pattern of neural activity tends to appear in the motor region of your brain. If you had a system to read your neural signals and recognize when certain cognitive events occur (like the imagined fist clench), you can use those signals to trigger a computer command. It’s like clicking a computer mouse, but instead of using your hand, it only requires thinking a specific thought!

Now, what good is a BCI and why do I want one? In the most extreme cases, BCIs can provide life-changing outcomes to patients. Neuralink’s recent clinical trials are a great example of this. Their BCI implants are showing remarkable improvements to quality of life in paralyzed patients, who are able to control devices that they would otherwise have no means of interfacing with. Even for those without paralysis, BCIs can provide new ways to interact with technology in daily life, as well as provide new research tools. In my case, I plan to connect my BCI to my digital audio workstation and explore new ways of creating music. I also have several completely novel research directions that I’ll cover in the future, so make sure to keep in touch if you’re interested in hearing more on these.

Measurement and a Challenge of Scale

In order to build a BCI, first you’ll need a way to measure brain signals. There are various approaches to measuring neural activity, but today we’ll focus on the most common method for BCIs, Electroencephalography (EEG). The nervous system communicates using action potentials, which are electrochemical spikes that run the length of neurons to alert neighboring cells when it’s time to fire. The sequential and combined firing of neurons is what drives behavior and cognition as we know it, making these electric patterns an excellent candidate for monitoring neural activity for our interface. EEG uses electrodes placed on the scalp to measure these signals.

Although it would be grand to measure activity of every individual neuron, there are far more neurons in the human brain than is practical to measure. The most recent estimates put the number around 86 billion neurons, and that’s not including non-neuronal cells like the glia. It’s currently impossible for us to monitor all of these cells at once, and even if we could, it would be difficult to meaningfully process a dataset that large in real-time. Just imagine the lag in your inputs as you try to control a cursor…

Instead of measuring all the cells at once, what if we monitored groups of neurons at the same time? As you might imagine for something so small, each individual neuron doesn’t produce that much electrical activity. However, if we have many cells activating all at once, their combined electrical activity may be enough to register across the scalp at our electrode. This is the key to EEG; since cognitive functions typically involve the coordinated activity of many neurons, we rely on this grouped activity of cooperating neurons to outcompete the noise generated by lone neurons.

Cleaned EEG signals. The smooth, less jagged curves suggest that neurons were firing in sync between the -4 and -2 time stamps.

Once we have our electric signal gathered from the EEG, we’ll want to decompose it to better analyze and understand it. We do this using a Fourier transform, which essentially lets us break down the complex waveform into its basic sine wave component. Neuroscientists have shown that activity in different frequency ranges tends to associate with specific types of cognitive activity. For example, delta waves with a frequency in the 0.5 - 2.5 Hz range are associated with deep sleep, while alpha waves around 9 - 13 Hz are associated with a wakeful but relaxed state. Decomposing these signals provides us with more targets with which to build our BCI classifiers, which are the algorithms we’ll use to recognize specific cognitive events.

So why is EEG so difficult to do at home?

To explain why I consider my successful home test an exciting development, we’ll visit a few stories from my PhD training. For a couple of years, I worked with a lab that specialized in brain-computer interfacing. I had a fantastic experience training with this lab, and through my work with them I learned a great deal about best practices and pitfalls of EEG. During this training, I identified two main challenges I would need to overcome to create an at-home BCI.

First, high-quality EEG equipment is expensive. Top-tier, medical grade EEG companies don’t share product costs except via direct requests, and I’ve heard of labs spending many tens of thousands of dollars to build out their EEG workflows. On the other hand, consumer-grade EEG products are available for hundreds of dollars, but I’ve spoken with researchers who have found the signal-to-noise of these devices lacking. Also, these consumer products can be tied up with proprietary software, which throws an extra layer of difficulty to the process of building your own BCIs.

A second major barrier for at-home BCI is electrical interference. The PI of the lab often told a story about his previous lab space, where his first EEG measurements went rather poorly. The experiments would go flawlessly in terms of participant performance, but when it came time to review the EEG data, it was a mess! Changing out the headset, the amplifier, even the experimenter… none of these eliminated the data quality problem. After much head-scratching, he eventually discovered that an unshielded wire ran through the center of the ceiling above the lab. The magnetic interference from the wire was strong enough to interfere with the EEG signals, rendering them functionally useless. He was able to move to a new lab space that did not have this level of interference, but even so, conducting EEG work in unshielded spaces is notoriously difficult.

EEG signals are so slight that they are difficult to detect. To assist with this, we feed these signals through an amplifier, like this one. This little guy is the OpenBCI Cyton + Daisy board, and can handle 16 channels of EEG data at once.

In order to build a BCI, both of these problems will need to be resolved. For the cost barrier, OpenBCI offered one potential solution. A full EEG workflow from them is still expensive, but at time of writing this, it is in the $2-3k range instead of $50k+. Also, most of their products are open source, so if you’re handy with circuits and 3D printing, you can save costs by building parts yourself. When I acquired my headset a few weeks ago (I’m using a gelfree cap with a Cyton + Daisy board), I wasn’t sure how the signal quality would compare. There are many published papers using OpenBCI tech, which suggested the quality is decent. Even so, I was amazed at how easy it was to capture high quality neural signals with my system. It’s worth noting that I have no affiliation with OpenBCI, I’m just very impressed that their tech has enabled me to overcome both of my biggest bottlenecks on this project so quickly.

Where to next, BCInterfacer?

Now that I have confirmed I can gather quality brainwaves at home, the next step is to build a classifier. Brain waves are highly variable, and in order to work with a BCI, you’ll need a well-trained classification algorithm to make sense of all that variance. To start, the goal is to make a classifier that can recognize and differentiate when I’m imagining clenching my fist vs not. With time and practice, this will expand to include many more neural classes, allowing the BCI to recognize a wider range of cognitive inputs and in turn, expanding the versatility of the interface. Once built and trained, I’ll be able to hook up this finalized BCI to various commands on my computer. From there, the bounds of human imagination and ingenuity are the limit!

—

As always, thank you for your valuable time, and keep an eye out for updates on this project in the near future! Many fantastic things are afoot…

Patrick Seebold, PhD