Thousands of Needles in Your Skull? MEAs and the Future of Invasive BCIs
You’re lying on an operating table. Your eyes flash open, the light hits quickly. As your eyes adjust, you find a mirror.
Where are you? What happened? Your feel a sharp pain all over your head like you’ve never felt before.
As your eyes adjust, you turn to look in the mirror, and slowly you realize: there are 10,000 needles piercing the deepest depths of your brain! How did this happen? Did you get abducted by aliens?
Turns out, this is your fault! The year is 2050, and you wanted to undergo the brain-needling procedure yourself, because you didn’t want to be left behind. Left behind from what? Let me clarify: These needles are the leading method of creating a brain-computer interface, an electronic structure of highly various kinds that receives and/or inputs electrical signals from/to the brain. I want to try to give you a window into this incredible piece of technology, so come along!
Quick Anatomy of a Neuron
To understand this, you’ll need to know some basic neuroscience (if you already do, feel free to skip ahead). Essentially, when neurons are ‘activated’, positive ions move into the cell body and negative ions exit. This is called an action potential: every thought you’ve ever had has just been an incredibly complex combination of action potentials across an even more complex network of neurons. Action potentials are ‘all or nothing”, meaning that either an action potential occurs or it doesn’t, there are no levels of strength for an action potential.
Methods of Brain Interaction
Some methods stay outside the brain (called noninvasive), like Electroencephalogram (EEG), which uses electrodes on the surface of the scalp to measure many many action potentials of nearby neurons: Because they’re on the surface, though, they can’t nearly reach the mysterious depths of our brain!
What about Electrocorticograms (ECG) that are surgically inserted between the skull and the brain. Trust me, you do not want to see a picture! These electrodes have more capabilities due to their immediate proximity to the brain, but remember, they’re still only on its surface. So although they can pierce a little deeper than EEGs, they’re still not reaching our full brain-probing potential.
So what if we actually gather data from and input data to each individual cell? Patch clamps connect directly to singular cells and are able to pierce deeply into the brain. They too are not without issues, though: they’re bulky and highly imprecise!
It seems like with each method, we have some major problem which severely limits our data-gathering ability. Can we create a technology that solves all these problems simultaneously? Yes, and their name is:
Multi-Electrode Arrays (MEAs)!
With MEAs, we get multiple simultaneous neuron measurements wherever, easy installation and setup, and selective stimulation of neurons (input instructions to neurons)! So, yeah, these guys are super useful! Now the question becomes how they work: to understand how the array works, as with most problems, it’s easiest to reduce the scale and eliminate everything but the core functionalities.
Basic In Vitro Microelectrode Array
Let’s start with an in vitro (not in a multicellular organism) array of 4 by 4 electrodes. On these electrodes, we’ll randomly place 8 pre-connected neurons.
When a neuron activates, an electrode converts the change in electrical charge from the movement of ions to an electronic current (ions vs. electrons). For a single neuron,
The voltage output of the electrode depends directly on the total change of ion charge in the overlapping area between electrode and neuron and the ratio of the overlapping region to the entire electrode.
Wow, that’s a lot! So basically, the voltage value that a given electrode outputs at a given time depends on the change in ion voltage it detects and how much it overlaps with the neuron. The more it overlaps and the larger change from negative to positive ions that the electrode detects, the higher the voltage output! So now we can get the instantaneous ion composition of each neuron in the form of voltage, which corresponds directly to whether the electrode is undergoing an action potential or not. But:
How do we determine exactly when it experiences an action potential?
We can set a floor for how many volts the signal of the electrode has to be to count as an action potential for its corresponding neuron. This floor is typically calculated as x times the standard deviation from the mean or median of the value of ion composition of neurons in time periods of inactivity, which still have electrical activity due to noise (x is often 4 or 5).
How do we actually read the data?
Each of the 8 electrodes has its own horizontal line which spikes when an action potential occurs. Scientists can then observe the patterns and structure of bursts of spikes, and make inferences based on these patterns. We’ll get into data processing later when we talk about actually used MEAs, which are more complicated.
So what if you wanted to use this technology to look at actual neurons still in living organisms?
Well, you can, in two ways. The more basic way is using an electrode for each neuron, just as in the in vitro array. Although this limits your ability to measure to less than about 500 neurons, the method yields more precise measurements which may be desired depending on the nature of the experiment.
Single-neuron measurement uses microwire arrays, advantageous because they can be bent and positioned precisely so as to reach a singular neuron and not be constricted by extraneous structures such as glial cells, who are more or less the caretakers of neurons. The data is received and analyzed the same way as in vitro arrays.
Now to in vivo!
Once we get to the neural networks of living structures on a macro scale, though, it becomes unrealistic to get data from each neuron among the billions in your brain (at least 100 billion, and potentially 50 times more glial cells!). How do we solve this problem? The solution that comes to mind most quickly would be to somehow measure multiple neurons with each electrode. That method happens to be the one we went with!
Let’s again start with the most basic form of this in vitro array, called the Utah Array (cause it was created in Utah, maybe the only thing Utah ever did!). A is what it looks like: a regular array of electrodes, except the electrodes are at the ends of incredibly intimidating needles so they can reach inside the brain. The only setback of Utah array electrodes is that they only receive electrical stimulation from the tip. To solve this, we use Michigan array electrodes, which allow for obtaining signals along the length of the electrode as well. Each electrode ends in an intercellular space (not in a cell).
So what exactly does each electrode measure? An easy way to gauge the aggregate of an area’s action potentials at any given time is to simply do the same thing as before: convert the change in electric charge from ion movement into electronic voltage, except this time, each electrode is further apart and measures charge in extracellular space. Therefore, the more negative the ions in the extracellular space, the more action potentials currently are occurring (since positive ions would have left the extracellular space to go into the cells).
The impact of currents traveling between neurons on the electrode’s final voltage output depends on the same factors mentioned above: magnitude of ion change, distance from electrode, noise, etc. In this way, each electrode can potentially get a reasonable assessment of action potentials for all the neurons within 100 micrometers of the origin of the action potential.
For each neuron, the voltage of the electrode is directly related to the point current value and inversely related to the conductivity of the medium and the distance from the point source of action potential to the electrode.
It’s important to not, however, that the change that the electrode detects is rarely the true change in ions. Numerous factors influence what the electrode detects, which are super important to know for the operation of an MEA. Let’s start with the most simple:
- The Medium of the Cell
Essentially, how well does the medium conduct the change in ions from the neuron to the electrode? This is determined most directly to the medium’s innate electrical conductivity: how easily a medium allows current to flow through. The quality of the medium is also measured by its Capacitance, the ability of the medium to store electric charge. Lastly, we have Homogeneity, which is a universal word meaning how similar things are to each other, here, specifically how all physical properties of molecules in the medium are the same. For our purposes, in the ideal cell, we would have high electrical conductivity, high capacitance, and high homogeneity.
2. The Electrode
The electrode itself, of course, influences the total voltage absorbed as well. The shape of the electrode is the most apparent. Then comes electrical impedance, an infamously difficult concept that boils down to a circuit’s resistance to alternating current (current that reverses direction periodically).
The electrode also needs to transfer the signal it receives to the reception location (central receiver). So, how well does it do that? Specifically, the factors here are data transmission, how well it transfers the data, and signal amplification, how large the signal is amplified.
Next comes the infamous factor of signal detection:
3. Noise
Basically, think of noise as any sort of interference that can influence the total change in ions detected. Noise can be biological, such as neural activity from the heart, or external, such as radio waves.
Some of the more prominent biological noises interfering with our neuron action potential detection are the action potentials of distant cells. Naturally, distant cells negatively affect the ionic composition of the intracellular environment, but when we aren’t trying to measure their impact, this can be less than useful. The heart is also a large interferer.
And, of course, the electrode itself generates noise as well, in the form of voltage and heat. There’s also the noise from the device that amplifies and digitizes the signal.
Power lines are also a famous factor. So, yeah, there’s a lot of noise! How do we counter all this interference? Well, to use the example of power lines, which emit waves at 50–60 Hz, we can just exclude that frequency in the data processing phase for this reason. Or, for the excess neuron noise, scientists use complex mathematics to isolate the signals of the neurons from which they want to get data.
What’s the purpose of all this?
The electrode can record the instantaneous ion constitution of the extracellular space, which accounts for the overall pattern of action potentials of the neurons in an area. This can be used to measure ‘brainwaves’, which are patterns of action potentials that reveal generalities about the subject’s state of consciousness, feelings, and intents shown as changing lines.
So what actually happens to change the raw voltage to usable data?
How to transfer the data
First, we want to separate the fast action potentials from all the noise we mentioned before, which we can do by applying a filter of typically 300–3000 Hz, a narrow band that significantly increases the accuracy of readings. After the signal is filtered, spikes are detected the same as before: an amplitude threshold.
Next comes spike sorting, which most often uses (buzzword incoming) machine learning algorithms! to group the spikes in activity according to their shape. In the ideal scenario, each neuron fits into one and only one cluster based on their spikes. We won’t go deep into this sorting stage due to its potential to be complex and variable, but if you want to know more about this aspect of MEA< there’s definitely research on there about different sorting methods.
How do we analyze all this?
So you have the data of spike grouping, now what do you do with it? Ideally, you can find out something useful about the subject based on the behavior they exhibit or state they are in, and vice versa. We can do this by focusing on bursts, when a neuron exhibits multiple action potentials in a short amount of time followed by a period of relative rest. Most especially useful is the focus on multi-neuron bursts.
Say you have a given simple action, like moving one’s left pinky. We can ask the subject to try to move only their left pinky multiple times (this works even if the subject's pinky is not there anymore). Every time they do so, we can now observe the sequence of action potentials in their motor cortex, which are slightly different each time, and use Machine Learning Algorithms! to create a trend of action potentials when the pinky fires. Then, every time the MEA detects that action potential pattern, you can activate, say, a robotic pinky to contract!
Here’s another one: an experiment with cats! 8 sets of electrodes were implanted in a cat’s parietal cortex during wakefulness and sleep state. During wakefulness, intracellular space was dominated by low-amplitude fast-frequency waves of charge. Transitioning to sleep, though, fast waves gave way to high amplitude slow waves dominated, but then in REM sleep gave way to the same low amplitude fast-frequency brainwaves exhibited during wakefulness! Interesting…. Perhaps because the parietal lobe is sensory and we dream during REM sleep?
MEAs are also capable of stimulating electrical signals, which I won’t talk about here. So yeah, they’re pretty amazing. I encourage you to explore the Brain-computer interface area further since it has a lot more to offer!
Please help me by sharing this article with anyone interested in neuroscience and/or emerging technologies. If you enjoyed the article, follow me on Linkedin. If you like this sort of thing, I think you’ll also enjoy my book, An Illustrated Exploration of Consciousness, about the mysteries of our mind.
Thanks for reading, I appreciate you!
