How to insert a neural implant like you floss your teeth

In this episode of the Neural Implant Podcast, Ladan interviews Dr. Ryan Clement about his work on a mechanism to make it much easier to insert neural implants. The trick? Move the implants back and forth very quickly, just like floss!

Tl;dr: In this post, I’ll walk through Ladan’s discussion with Dr. Clement about NeuralGlider, a device that helps researchers insert neural implants without damaging brain tissue. Then, I’ll discuss the development of medical products like NeuralGlider and how the development process is impacted by regulation. Finally, I’ll let my thoughts wander to the regulation of software products.

The Episode

Dr. Ryan Clement is the pre-clinical studies leader at Actuated Medical, a company that develops medical devices relying on ultrasonic vibration technology. According to their website, “Ryan directs our testing program ensuring it is verified, validated, and generates the scientific data required for regulatory clearance.” Simple enough!

Actuated Medical doesn’t only develop devices that fall within one particular domain or use-case; rather, they take underlying expertise in ultrasonic technology and look for use-cases where ultrasonic technology is uniquely useful. In the past, they’ve made devices such as TubeClear, which can clean clogged feeding tubes through vibration. NeuralGlider, in contrast, applies ultrasonic technology to brain implants.

Image: NeuralGlider. This is an image of the NeuralGlider insertion device.

Let’s talk about the NeuralGlider device! The patent for NG can be found here, in case you feel like getting mired in dense legopatentalese (just coined it; “legal patent language”). The idea is as follows:

When electrode arrays are implanted, they can dimple the brain tissue. This is undesirable because this dimpling damages tissue and can reduce the quality of electrode recordings, either due to biological reaction or due to damage to the probes (electrodes) themselves. The nature of the dimpling effect is that when you pack more electrodes tightly together, the dimpling becomes worse.

Let’s build some intuition for this. Suppose you’re on a trampoline, and because you’re a phenomenal gymnast, you can do a one-armed handstand. When you’re in your one-armed handstand, the trampoline deforms downwards significantly; the pressure applied is high (large force, distributed across the small surface area of your hand). Now, suppose you put all four limbs on the trampoline: there will be a dimple around each hand and foot, but the dimple will be much less deep than when you were doing your magnificent one-armed hand-stand. This is because the force is distributed across all four limbs, so the pressure applied by each limb is smaller than when you were doing a one-armed handstand.

In the case of a trampoline, where you really want the trampoline to stay intact, this is perfect. Distribute the force! But, neural implants are another story altogether. The electrodes need to poke through the pia, the layer surrounding the cortex. If you insert one electrode, it applies force to a small surface area, therefore making it easy(ish) to rupture the pia and insert the electrode into cortical neurons. However, if you have a whole array of electrodes (equivalent to a whole lot of hands and feet pressing into the trampoline), each individual electrode/hand/foot is applying very little force to the pia…and therefore, in order to actually penetrate the pia, you need to displace the tissue further (therefore applying more force). In other words, you’re on all fours on the trampoline, and you need to get a friend to sit on your back.

According to Dr. Clement, researchers have tried a couple of solutions. The first is to use pneumatic insertion, as in the case of the Utah Array. Essentially, a pneumatic cannon launches the array into the cortex; the velocity helps the electrodes poke through the pia without deforming the brain tissue.

Image: Rousche and Normann, 1992; a schematic for a pneumatic inserter.

Another option is to insert the electrodes very slowly, and hope that this will give the tissue time to stretch out and become easier to poke through. Still, Dr. Clement says, researchers might have to push the electrodes in 50% further than originally intended in order for the “brain to pop” (his words).

Actuated Medical’s solution is to rapidly vibrate the electrode array, making it easier to insert. You might be able to gain some intuition for this by thinking about flossing. Trying to nudge static floss between tightly-packed molars is extremely difficult; but, if you jiggle the floss, it becomes easier. Why is this the case? In physics, kinetic friction is lower than static friction (kinetic = the friction-receiving object is moving; static = the friction-receiving object is stationary). The moving floss experiences kinetic friction, whereas the still floss experiences static friction. In a similar manner, a vibrating electrode array experiences kinetic friction against the brain tissue instead of static friction, and therefore it’s easier to penetrate the pia.

Don’t believe me! Believe the data:

Image: NeuralGlider; “Actuated” means the implant was inserted using ultrasonic vibration; “Un-Actuated” means the implant was inserted normally. In these parallel images, it’s easy to see how much less Actuated insertion deforms the tissue than Un-Actuated insertion.
Image: NeuralGlider; “Agar” is a gelatinous model for brain tissue; “Rat” means a rat brain was used; “Porcine” means a pig brain was used. In these graphs, the y-axis shows the percentage of the Control. So, the gray bars (which display Control data) all have height y = 1 (100%). The orange bars, then, show what percentage of the Control force/surface displacement the Vibrated electrodes cause. The orange bars are very short relative to the gray bars; this system works pretty well!

Rumination

The introduction of neural interfaces into therapeutic and eventually enhancement settings will be a long process. One interest of mine is looking at the dynamics of the BMI/bioelectronic medicine world from a high level, and eventually using that knowledge to help craft ethical and effective paths forward. As such, it’s important to understand the neural implant development cycle at all levels of abstraction. Here, we’ll study the level of an individual product developed by an individual company, based on some great information on Actuated Medical’s website.

Actuated Medical is a medical device company, and developing medical devices is obviously different from developing e.g. software products (which is the realm of technological product development with which I’m the most familiar). Let’s first build an understanding of what Actuated Medical’s particular medical device development cycle looks like, and then we can compare that to software development. To do so, we’ll use a case study from another one of Actuated Medical’s products, called TubeClear. TubeClear is a device that can clear a clogged feeding tube—a feeding tube is used to deliver nutrients to a patient who can’t eat due to their condition or treatment. (No, this isn’t a neural implant — but, it is a medical device, and therefore subject to similar considerations and regulations, so I think it’s a reasonable proxy for discussing neural implant development.)

Actuated Medical’s website specifies their particular seven-step process of developing products (the following words are quoted directly, but the list formatting is my own addition in order to improve readability):

1. Clinical need was understood
2. State of practice was identified
3. Clinical consultants were selected
4. Proposal was written and submitted in December 2007, resulting in a Phase I Small Business Innovation Research (SBIR) grant from the National Science Foundation (NSF) — awarded in July 2008
5. During Phase I, our device was investigated (Stage 1) and proven feasible (Stage 2)
6. We submitted a Phase II SBIR grant proposal in December 2008, which was awarded in July 2009
7. The device was developed from concept to verification-and-validation testing (Stages 4 & 5), and a 510(k) application for market clearance was submitted to the FDA June 2011

Here, the “Stages” refer to Stages of Actuated Medical’s development cycle. We can also take a look at a visualization of their product cycle, found on their website:

Image: http://actuatedmedical.com/approach.html.

(A quick note: the FDA website says that TubeClear was given approval in December 2012.)

Let’s dissect some of the similarities/differences between this product development pathway and the product development pathways you see at technology companies.

Need-finding: same!

No matter what product you’re developing, you need to first identify a need for that product. Otherwise, no one will buy it. If you ever find yourself living in Silicon Valley, you’ll hear this over and over and over…and over again, probably to the point of sigh-inducing banality.

Identifying the state of practice: similar, but there are differences

I’m not privy to the details of Actuated Medical’s state-of-practice-identification process, but the idea seems rather self-explanatory: figure out what solutions are currently used to e.g. clean clogged feeding tubes, and determine whether or not a better solution is desirable/imaginable.

The analogue in technology development is performing competitor research and understanding solutions offered by competitors (for example, if you’re building Dropbox, you would likely want to research Google Drive and figure out what its strengths and weaknesses are). Although I’ve never been part of a medical device’s development, I’m imagining that “practice” for medical devices doesn’t just entail “what does that piece of software do?” It probably implicates: insurance, regulation, materials, medical risk, cost, training practitioners to use the device, maintenance/upkeep…etc. In other words, I think it’s probably a more complex task to understand the current state of practice for a particular medical use-case than it is for a particular software use-case.

Grant funding: different!

This is a major difference between most software technology and medical technology businesses: MedTech business can apply for grant funding from the government. Unlike venture capital or other forms of investment, this grant funding a) is non-dilutive (which means the investor doesn’t gain ownership over some portion of the company), and b) doesn’t need to be paid back. Actuated Medical, in particular, pursues Small Business Innovation Research (SBIR) grants.

A quick aside on SBIRs

SBIR grants are government-given grants targeted at for-profit businesses with fewer than 500 employees. Each year, $2.5bn in grants are awarded through the three-phase SBIR program.

In Phase I, companies are awarded up to $150k for six months of exploring the technical merit/feasibility of any idea.

In Phase II, companies are awarded $1M for up to two years to explore commercialization, assuming positive results from Phase I.

In Phase III, no additional funding is given; this is when the business takes the product to market and must rely on other funding sources (such as debt or venture capital).

One large benefit of the SBIR program is that the companies own their IP (intellectual property) and commercialization rights to the technology they develop with grant funding; as such, some well-known and extremely successful companies obtained SBIR grants in their early days. Examples are Symantec (cyber security), Qualcomm (semiconductors and telecom), Da Vinci Surgical Systems (robots for performing minimally invasive surgery), and Jawbone (fitness trackers).

Market clearance: different!

“Market clearance” means that the FDA has given a company the thumbs-up to actually sell their product. This is another enormous difference between MedTech and software technology. Non-regulated software technology can launch whenever…this was the case for my previous company, Altar. (Note: some software technology is regulated, such as software that works with health care records, or software that deals with financial transactions).

In contrast, medical device regulation is extremely complicated, and people make entire careers out of consulting with companies who are trying to navigate the FDA’s regulatory process. To give you the tip of the convoluted iceberg:

A quick aside on medical device regulation

There are three federally-specified classes of device: Class I, II, and III. Class I devices are the least regulated, and Class III are the most regulated.

When developing a Class I or Class II device, a business is required to tell the FDA before it starts marketing and selling the device. This is known as a 510(k) premarket notification. As part of the application process, the company must prove that their device works similarly, is similarly safe, and is similarly effective to another device which was already given the thumbs-up under 510(k). This is known as “substantial equivalence.”

If a device fails to obtain 510(k) approval because the device isn’t substantially equivalent to another device which came before it, then the company can make a De Novo submission and request their new device to be classified as Class I or Class II (Class I & II devices are devices that obviously need to function correctly and be safe, but which aren’t directly sustaining or supporting human life). If the device does directly sustain/support human life then the company must make a Premarket Approval Application (PMA) for the device to be classified as Class III. PMAs are the most stringent premarket review process specified by the FDA.

In general, navigating the FDA regulatory process is extremely complicated, and requires years, millions of dollars, and trained consultants. This is the biggest difference between medical technology development and unregulated software technology development.

Pivoting the Rumination to a Different Topic

While writing this, I couldn’t help but wonder what regulation might look like for consumer-facing software applications. So, here we go…

I think there could be two broad categories of regulation: data privacy and user health. There’s certainly discussion in the public forum around the former, and it’s reached a peak lately due to the Cambridge Analytica/Facebook fiasco. The demand for this regulation is high, and the difficulty of implementation is manageable (c.f. HIPAA, a piece of legislation designed to protect the privacy of digital health data).

More thorny and amorphous is user health, or in other words psychological impact. It’s undeniable that consumer-facing software technology can have negative psychological outcomes; this is particularly true of social media. The intuition can be found in the compulsion some (*raises hand*) feel to check their (*my*) phones. We seek recurring rewards: every time I open Facebook, I’m hoping for a red notification symbol, and when it’s there I feel the same as while I’m anticipating it. If there’s no notification, I feel a little hammer chipping away at my soul (and my dopaminergic neurons probably decrease their firing rates significantly). An interesting life-hack I recently came across is (if you have an iPhone) to switch your phone into black/white mode. Give it a try, tell me what you think…

There’s also science here, but it unfortunately hasn’t entirely converged. There have been conflicting results, but studies like this and this are reporting negative associations between Facebook usage and well-being. This paper reviews the psychology literature on social media and addiction.

When a consumer-facing technology product is nascent and has very few users, I would argue that user-health-focused regulation is irrelevant. But, when a product has a usage magnitude on the order of millions or billions (echem, Facebook), then it’s time to start talking about the health of users. The notion of well-being-focused regulation is…difficult. There’s huge variability in how people respond to technology platforms (my evidence for this is anecdotal, but my anecdotal sample size is roughly the size of my real-life social network). Furthermore, the line between “this person isn’t exercising the willpower to not overuse technology” and “this technology has caused this person to become addicted to it” is extremely fine, mired in free will, and generally finds itself tiled with very sharp conversational thorns.

I’m proposing the following regulatory framework not because I’m for or against it, but for the sake of a thought experiment regarding the human impact of technology. Here we go:

A proposed software regulation paradigm

Not everyone who uses a given piece of software (let’s say, Facebook) will become addicted or experience a large amount of negativity due to using it. Let’s say that Y% of all people are susceptible to this negativity, and we decide that if more than Z people using a given platform are susceptible to negativity, then we ought to do something about it. We can use this to calculate a “threshold of alarm,” denoted T. This is calculated: T = Z / (Y / 100) = 100Z / Y. If we decide that no more than Z = 500 users can be susceptible, and Y = 4% of people are susceptible to this given platform, then our threshold is T = 500 * 100 / 4 = 12,500 users.

(For the sake of this thought experiment, I’ll gloss over the problem that I think is most immediate and deathly glaring: how on earth do you find a reasonable susceptibility metric for every consumer software product on the market?)

When a software product launches, assuming its susceptibility factor (= Y / 100) is known, it wouldn’t yet need to be regulated. This would allow the classic speedy and iterative development paradigm Silicon Valley holds so near and dear to its (Silicon?) heart. However, upon surpassing the threshold of alarm (T), the company could be required to enter into a software-focused version of “clinical” trials, where:

  1. The positive values of the software product are clearly articulated and tested according to the level of rigor held as standard in academic psychology or neuroscience circles; in essence, the question of “is the reward worth the risk?” is asked.
  2. “Side effects” (i.e., negative psychological or physiological effects like heightened blood pressure) are characterized.
  3. If the results of the previous two inquiries aren’t up to some standard of operation, then the company would be required to modify the product over some time period in order to continue growing the user-base. The goal of product modification would be to lower the susceptibility factor and maximize the positive value of the product.
  4. Even without offering any detail on implementation, those previous three points sound marvelously tedious. So, this sort of regulation would place pressure on a company, from Day 1 onwards, to build products that have clearly articulated human impact value, in addition to the requisite economic value. This is the most important part of the hypothetical regulation, in my opinion: biasing companies towards being conscientious of their human impact.

Although just a thought experiment where I egregiously gloss over details, I think this is nonetheless worth proposing. In case I failed to make it apparent through my writing, this line of thought derives from my philosophy that technology has to be designed to improve the human experience, and minimize unintentional fallout from doing so.

In Conclusion

In this post, we discussed the ultrasonic neural implant insertion apparatus pioneered by Actuated Medical, called NeuralGlider. We then studied Actuated Medical’s case study of their own product development process, and compared medical device development with software development. Finally, we wandered down thought experiment lane in an effort to imagine what regulation of consumer-facing software technology might look like.

One of my primary goals with The Substrate is to encourage an ethics-first conversation about brain computer interfaces. To that end: comment, write me at hello@thesubstrate.com, or find me on twitter (@averybedows). I will respond! (seriously, I will).

Until next time,

Avery