From the Interface's RSS Feed

Science Corp and the quest for a Vision BCI

Thu, 30 Oct 2025 00:00:00 GMT

The success of the PRIMA implant in clinical trials makes Science Corp the leader in vision BCIs. The company is also simultaneously pushing research and engineering forward along several other fronts, all while staying focused on retinal prostheses. This post tracks Science Corp’s progress since it was founded in 2021 by Max Hodak and other ex-Neuralink engineers.

Success of the PRIMA implant clinical trial

The positive results from the pivotal clinical trial of the PRIMA implant were recently published in New England Journal of Medicine, showing that the implant restored vision in patients with age-related macular degeneration. Importantly, this is the first vision prosthesis to provide patients with ‘form vision’ that allows them to recognize letters and read. Previous attempts at vision neural interfaces showed that the brain could be stimulated, but in response to visual stimuli patients would see ‘phosphenes’ or flashes of lights that were not correlated to the shape or form of the objects that they intended to see. The 2X2 mm, 378-pixel PRIMA implant also has the advantage of being wireless and powered through the same laser light that stimulates it.

Still from an video with results of the PRIMAVERA clinical trial showing a patient filling out a crossword puzzle using the implant.

Science Corp acquired the PRIMA system in April 2024 for €4 million along with other assets and staff from the French company Pixium. Pixium in turn had licensed the technology in 2013, originally patented by Daniel Palanker and team at Stanford in 2004. In all, that makes over 20 years of development cycle from the lab to being used in patients — a cycle that the Science team and other BCI companies will hope to shorten for future devices.

The 2mmX2mm 379-pixel retinal implant and the glasses with the camera-infrared projector and camera. From the Science Corp website.

The Science Eye and the biohybrid implant

Science Corp’s first announcement was the ‘Science Eye’, a supra-retinal implant that delivered stimulation to the optogenetically modified retinal ganglion cells (RGCs) through a micro-LED display panel. The chip had an extremely high resolution for stimulation compared to the PRIMA, but the RGCs are further downstream (towards the brain) than the bipolar cells that the PRIMA system targets, and it hasn’t been shown that stimulating the RGCs can lead to ‘form vision’. Crucially, optogenetic modification has never been done in the human brain cells, so this implant would have been decades away from clinical use, for safety reasons. So it’s not surprising that they chose to prioritize PRIMA which is much closer to clinical trials and human use.

Microscopy image showing axons (green) from the implant deep into the cortex(blue nuclei). Image from the preprint paper.

However, the company is still betting on optogenetics through their work on the biohybrid implant. In a 2024 preprint describing their research on the biohybrid probe, they proved the concept that neurons grown outside the body could be made to communicate with brain cells. The study allowed the team to conclude that mice implanted with a biohybrid implant were able to respond to the optical stimulation of the implanted neurons. Although the study had several caveats — the change in behavior in the mice was not statistically significant; the optical stimulation was extremely strong; and there was no recording of activity originating from inside the brain indicating a 2-way interface — the concept is promising and is likely to make its way to humans before technology that requires directly modifying human brain cells to respond to light.
The neuron-chip interface in this biohybrid implant device is outside the brain with axons growing out to form synapses with existing neurons in the brain. Recording and stimulation takes place outside the brain and can be observed and measured and controlled better, with much less damage to brain tissue, which should enable a much higher bandwidth and a longer lasting implant. This is not a new idea, so it's surprising that none of the other established implantable BCI companies have gone this route yet.

Automated channelrhodopsin engineering

Science Corp’s newest preprint describes work that is even further upstream in the BCI R&D pipeline. The key innovation was to automate and scale up the development of new channelrhodopsin variants. Channelrhodopsins are the proteins that make neurons sensitive to light so that they can be stimulated as part of a neural interface. Science Corp’s team developed an automated high-throughput screening method that they could use to select channelrhodopsin variants with promising properties. In this case they selected one that was sensitive to low levels of light — a property that could feed into future versions of their implants and can perhaps be commercialised independently. The automated protein engineering technology that allowed them to run this screen is equally valuable as an asset, and could potentially be developed into a product for other researchers, or run as a service for the same set of customers that use their MEMS fabrication service.

MEMS fabrication and readymade ASICs

Science Corp’s ‘foundry’ is a MEMS fabrication facility, offering researchers to outsource the fabrication of semi-custom chips that they can use to run experiments. MEMS stands for micro-electro-mechanical systems. These are the chips needed for research and development of almost any kind of implantable neural interface. Setting up a MEMS facility is extremely capital-intensive and it requires specially trained staff. Science Corp acquired a MEMS Facility in 2022, and has since been running it as a commercial shared service for like-minded labs both in academia and industry. While they likely use a fraction of its capacity for their own projects, allowing customers to share the costs accelerates innovation in the space. Precision Neuroscience is another BCI company that has a foundry and offers MEMS development, with a similar business model.

Finally, the startup has some off-the-shelf products for labs that work in similar areas. These include programmable ASICs for optical stimulation and electric recording, a headstage, and a software toolkit that works with their system.

The Science Corp portfolio

Science Corp’s portfolio of assets.

The company describes itself as being ‘fully vertically integrated’. An alternate way to phrase it is that Science Corp has a portfolio of investments in BCIs and BCI-adjacent neuroscience research at various stages in the pipeline. Their position is unique, and hard to compare their trajectory to the more established neurotech companies because each of their bets has a slightly different business model and a vastly varying time horizon to maturity. The programmable ASICs, the high-throughput channelrhodopsin screen and the new WiChR F240A channelrhodopsin are assets that they could commercialize early via neuroscience labs and startups, while also using the technology to accelerate their implants. The biohybrid probe is a very early prototype and it will likely be a decade or more before this comes to human trials.

Since being founded in 2021, the company has raised $290MM in total, with the last $100M being from a funding round in early 2025 on the back of the positive results on the PRIMA pivotal clinical trial. Science Corp is responsible for bringing the PRIMA device to patients in the EU, now that the pivotal trial was successful, while attempting to get approval for it through the US FDA. If that works out, the product could potentially generate enough revenue (or attract even more funding) to see their longer-term bets through to fruition.

Venture Capital Funding for Brain-Computer Interfaces outstrips DARPA dollars

Thu, 09 May 2024 00:00:00 GMT

A group of 5 implantable brain computer interface (BCI) startups, including Musk’s Neuralink, are starting to pull in hundreds of millions of venture capital dollars. While Neuralink has attracted the most funding by far, it’s not the furthest ahead on human trials or FDA milestones.

The neurotechnology giant Blackrock’s $200 million fundraise from Tether in April 2024 marked a turning point in the way that brain-computer interfaces are funded. BCI research has been driven largely by DARPA, but venture capitalists are now starting to invest bigger sums into a field that will likely take decades to pay off. At the same time, the United States Congress cut the budget for NIH’s BRAIN initiative, indicating that more Neuroscience and neurotechnology research will need to be privately funded moving forward.

Funding

Neuralink has always been privately funded with a total of $680 million, but most BCI companies have benefited from government funding through various DARPA or NIH grants. Musk is credited with bringing attention and funding to the field and raising the funding numbers by an order of magnitude, to the benefit of other manufacturers, who have made more progress with human studies.

There are 3 other major players in invasive BCI. Paradromics, founded around the same time as Neuralink with $88 million raised, has not started human studies despite an FDA breakthrough device designation. Synchron, originating in Australia with $145 million funding to date, has the most patients in clinical trials after Blackrock. The newest entrant, Precision Neuroscience, makes a device that lies on the surface of the brain rather than within it and has already been implanted in humans for safety testing and data gathering.

FDA and human studies milestones

Most manufacturers follow the following sequence: they petition FDA for breakthrough device designation. This optional step puts them in a program that speeds up communication with the agency, but it doesn't mean they're approved to start clinical trials or for their devices. When they're ready to start human trials, usually after animal testing, they apply for an investigative device exemption (IDE) from the FDA. This means they're authorized to begin clinical trials in humans. Initial trials must be small, primarily aimed at establishing safety over both the short and long term. These trials can take years to complete, with Neuralink's PRIME study spanning 6 years. Manufacturers often use these safety trials to demonstrate efficacy or some level of functionality, such as controlling a keyboard or mouse, or to gather data. Finally, most clinical trials are registered at clinicaltrials.gov for transparency (see Synchron’s COMMAND trial for example) so that the public can track the goals of the trial, the milestones, and progress — it also obliges them to report outcomes. Neuralink notably hasn’t registered their clinical trial.

Chart showing funding, FDA milestones and human clinical trials for the top 5 implanted BCI makers in 2024. Neuralink has raised the most funding but isn’t the furthest ahead on human clinical trials.

Blackrock

Blackrock’s 128-electrode Utah Array has been around since well before most people heard of brain-computer interfaces. While the other companies have one primary device that they are taking through the regulatory steps, Blackrock has taken a more pipeline approach. They have produced the ‘majority of brain implants that have been in humans’, and their BCIs have been through a variety of clinical trials at research labs around the US demonstrating control of motor functions through typing, moving prosthetic limbs, and translating sensations back from a prosthetic to the brain.

The Utah array was first implanted in humans in 2004, but the company Blackrock was founded in 2008. Until recently their BCI’s were designed for use in the lab as part of a research study, but in 2022 Blackrock reported a partnership to expand these clinical trials to a portable in-home BCI system ‘Moveagain’, which received breakthrough device designation in 2021. In 2022, Blackrock also announced ‘Neuralace’, a 10,000+ channel array on a flexible chip, that will become available to the neuroscience research community in 2024.

Blackrock has received significant public funding, both directly through grants, and indirectly through funding for research trials at academic hospital systems across the US. Their revenue for most of the company’s life has been from an array of research devices and systems for scientists to study mice, primates, and human diseases like epilepsy. After a smaller ($10M) round of private funding in 202, this new funding round from Tether will make the stablecoin firm a majority shareholder in Blackrock.

Neuralink

Neuralink’s story is well-chronicled. Founded by Elon Musk and a team of seven scientists and engineers in 2016, the first $100M came from Musk. Over the years the BCI firm has raised a total of $680 over 5 rounds from private investors, with the last round being $323 million in August 2023. Neuralink’s valuation is said to be in the billions, and the company has received no public funding.

Neuralink’s premise is its N1 Implant that records neural activity through 1024 electrodes distributed across thin and flexible 64 threads, and the use of a surgical robot to do the implantation because these threads are too fine to be placed by human surgeons. Neuralink secured breakthrough device designation for its brain implant in July 2020 and the company demonstrated a macaque playing pong with its brain in 2021. The company received an IDE (or permission to start human trials) from FDA in May 2023, after a rejection in 2022. Their PRIME study started recruiting September 2023 and will take 6 years to complete. Neuralink did not register the study at clinicaltrials.gov, but released a video in March 2024, showing a patient using the brain implant to play chess, followed by a 100-day update in May.

Paradromics

Founded in 2015, Paradromics has raised a total of $88.7M in private funding over several rounds, in addition to approximately $18M in research funding from the United States Defense Advanced Research Projects Agency (DARPA).

Their implant is a ‘high data-rate brain computer interface’ designed for long-term daily use. Paradromics received breakthrough device designation for its Connexus DDI device in May 2023, and the team aims to launch its first human clinical trial in 2024.

Synchron

Synchron was also founded in 2016 to acquire a spinoff from the University of Melbourne. The company has raised a total of $145 million in funding over 5 rounds, with their latest $75 million Series C round in December 2022. Synchron is backed by some well-known names including Bill Gates, Jeff Bezos, and Khosla Ventures.

Unlike the other BCIs, Synchron’s device - the Stentrode - is intravascular, meaning that it sits inside a large blood vessel rather than directly in contact with the brain. This means that it is safer to implant, but also perhaps limited in its possible location and consequently in its function. The Synchron Switch™ BCI (a platform that builds on the Stentrode device) received FDA Breakthrough Device Designation in August 2020.

Synchron’s first human study with 4 patients in Australia completed in 2022. The technology was not only safe, but also allowed patients to do hands-free texting, emailing, online banking and shopping, and communicating care needs using their thoughts. Following this, they started the COMMAND trial in 2022 at three US clinical sites. The primary goal is to establish safety, but the trial has also been designed to evaluate how the BCI may enable the use of patients’ thoughts to control digital devices for daily tasks.

Precision Neuroscience

The newest addition to this list of BCI makers is Precision Neuroscience. Founded in 2021, three members of the Precision crew are ex-Neuralink, including Ben Rapoport, who was the surgeon on Neuralink’s founding team. The startup has raised $53M in funding - $12M in 2021 followed by $41M in January 2023.

Their device is a thin-film microelectrode that sits on top of the brain rather than within it - a flexible, modular and bidirectional micro-ECoG, that is coupled with a minimally invasive “cranial micro-slit” technique for insertion.

Precision Neuroscience received a breakthrough device designation in 2023. Later that year, they also did a pilot study recording neural signals in patients undergoing neurosurgery. Their device was only temporarily positioned over the brain - a proof of concept that recordings were possible and data could be gathered. The company has not yet received an IDE, performed clinical trials, or demonstrated safety or efficacy.

Many of these companies, notably Neuralink and now Blackrock, want to one day expand the use of these implanted devices beyond cursor control and motor control to vision and to higher-bandwidth communication. Given that they are only now performing their first safety trials with single-digit patient numbers — this could be decades away.

Beyond Handwriting: Speculating on what is next for Ctrl Labs' Wristband Neural Interface

Tue, 02 Apr 2024 00:00:00 GMT

Ctrl Labs within Meta’s Reality Labs division released a paper about their incredible progress on a generalizable wristband-based generalizable neural interface. The choice of a handwriting demo task over the previously teased typing task is curious, and unless there is more coming soon, there is still a lot of work to be done before they can launch it as a product.

Soon after Mark Zuckerberg alluded to a neural interface that was was ‘... actually kind of close to…a product in the, in the next few years…’ on the Morning Brew show, the Ctrl Labs team at Meta’s Reality Labs released a research paper about their sEMG wristband.

The paper describes how they used data from thousands of participants to train models that allow their wristband prototype (or research device) to work out of the box for new users — a first in the neural interface field. Their research is robust and reported in detail, describing how the team

Combined their multielectrode sEMG bracelet hardware with a scalable data collection infrastructure
Used this setup to this to collect data from 1000s of participants for wrist movements, gestures like thumb and finger taps, pinches and swipes, and handwriting tasks
Developed models that achieved close to 90% classification accuracy for held-out participants on gesture detection and handwritten character recognition

Still from a supplementary video uploaded with the paper

They showed that:

Training on a large enough dataset allows you to build generalizable models for wrist sEMG (although the dataset itself isn't available).
Model personalization further improves performance and reduces latency.

The 3-page list of contributors indicates how much effort has gone into this since Ctrl Labs' acquisition in 2019. The demonstration of a neural interface that works without calibration is certainly noteworthy for the BCI world. However, there is still a significant gap between this paper and some earlier demos and teasers. In particular, their choice of a handwriting task, which achieves at best less than half the speed of typing, raises more questions.

I'd love to understand why they didn't pick the typing task they had previously shown in a video. In the paper, the authors stress that their ground truth was approximate and relied on prompts and inferred timing. A typing task — perhaps with a touch keyboard — would have provided them with true ground truth that was more scalable. Typing would also provide much more open space for adaptive learning, envisioned in Meta’s March 2021 post, “...imagine instead a virtual keyboard that learns and adapts to your unique typing style (typos and all) over time..”.

Still from an ealier video with a virtual keyboard

Other areas that Meta had discussed earlier, but this paper didn't touch on, were

barely perceptible controls or the use of much subtler non-perceptible movements (accessible neuromotor information that isn't being utilized),
intention and co-adaptive learning
using language models to correct human text.

Meta has previously demonstrated and discussed these, with evidence that they have research devices and models capable of performing these tasks. Hopefully this recent paper is just a start, and there's more coming soon.

One step closer to a brain-computer interface: Generative AI decodes images and speech from fMRI recordings

Sat, 03 Jun 2023 00:00:00 GMT

This post compares two new research papers that use generative AI techniques to ‘read’ images and text from human brain activity patterns. The first paper generates sentences that subjects heard in podcasts, and the second reconstructs visuals that they were shown — both using only fMRI readings. Together, they show that generative AI techniques can compensate for some limitations of fMRI to capture not only the literal words or pixels perceived but also their semantic meanings.

While these papers were formally published just after generative AI entered the mainstream consciousness through image generation from text (Stability AI, Midjourney, DALL-E), and large language models (ChatGPT), both were conceived in 2022 using older versions of these models. In many ways these papers are variations on the famous ‘I can see what you see’ era work from 2009-2011, with the latent diffusion and large language models now doing the heavy lifting.

Reconstructing text

In Semantic reconstruction of continuous language from non-invasive brain recordings, the researchers train an fMRI-to-paragraph generator for 3 participants, using fMRI data created while subjects listened to 16 hours of podcasts. Their decoder is able to capture the gist of the sentences at a level higher than chance, but the examples in the paper (image below) give the impression that the LLM is relying on its vast training corpora to to autocomplete a faint signal.

Reconstructing Images

The authors of Stable Diffusion with Brain activity use the open source Stable Diffusion model to reconstruct images from the Natural Scenes fMRI Dataset. They trained models to map from fMRI signals corresponding to the visual and semantic regions of the brain to the image and text components of the latent diffusion models. The results are spectacular, showing that the latent diffusion model combines the two components to produce images that are similar to the original both in meaning and in appearance.

Comparing the papers

	High-resolution image reconstruction with latent diffusion models from human brain activity	Semantic reconstruction of continuous language from non-invasive brain recordings
Published	CVPR 2023	BioRxiv Sep 2022, Nature Neuroscience March 2023
Generative AI model	Stable Diffusion	Original GPT
Key contribution	Reconstruction of perceived images from fMRI with much higher semantic and pixel-wise fidelity than previously possible.	First reconstruction of continuous language from fMRI. Prior work used a fixed vocabulary.
Individual-specific algorithm training	Yes, all models were built on a per-subject basis	Yes, trained separate model for each participant
Brain imaging modality	fMRI	fMRI
Dataset description	Used the Natural Scenes dataset of fMRI measurements of 8 healthy adults shown images from CoCo.	fMRI recordings from 3 subjects while listening to listened to 16 hours of podcasts
Dataset availability	Openly available through Access Agreement	Not openly available
Code available	Soon	Partly

These recent papers bringing AI and neuroscience worlds closer together were only possible because the 2022 versions of generative AI models were open access (like the Natural scenes fMRI dataset). If we had better neural interfaces (higher bandwidth, more fidelity and better temporal resolution) the models could do a lot more.

Wrist-worn neural interfaces have come of age: Facebook Reality Labs and Cala Health

Wed, 06 Oct 2021 00:00:00 GMT

Neural interfaces may be mainstream before we know it, in very familiar form. Facebook's Reality Labs division is working on a consumer device that senses tiny nerve signals at the wrist to offer unheard-of degrees of hand control. Cala Health has taken wrist interface technology into medical device territory. Their prescription-only wristband for individuals with movement disorders influences the brain's tremor network by stimulating nerves at the wrist. This blog post explores how these devices sense and stimulate nerve fibers at the wrist to create new pathways to and from the brain.

Futuristic devices that read signals directly from neurons in the brain have captured media and investor interest, based on the premise that we could one day communicate with computers at much higher speeds. The promise (or the threat, depending on your point of view) is that this 'neural lace' will allow us to mesh more closely with machines, to have them read our thoughts (and maybe write them!?) so that we can transcend our human limitations.

The technology that ushers in this sci-fi future will probably be much more prosaic — and safer — than Neuralink's invasive electrodes or even Kernel's TD-fNIRS helmets. Facebook's 2019 acquisition of CTRL-Labs and their continued investment in the wrist-worn nerve sensor technology mean that the first directly sensed consumer neurons will be those that innervate the forearm and hand.

Sensing from wrist nerve fibers just underneath the skin

The name EMG or electromyography sounds like it means electrical signals from the muscles — but what the sensors really pick up are action potentials from individual motor neurons that innervate these muscles. These signals are amplified and converted to detectable waveforms by the muscle fibers before the muscle contraction occurs. The sensors on the wristband pick up the intent to move, rather than the motion itself. This allows the user to have new dimensions of control and creates a faster, higher-bandwidth connection with the environment than our natural hands.

The wrist is the perfect candidate for a consumer neural interface because it allows for non-invasive access to nerve fibers running just beneath the skin. Wrist EMG is physically comfortable, and the form factor is familiar. More importantly, it's reassuring that these devices can't really tap into your thoughts but can only intercept the volitional control of your muscles.

Facebook/CTRL-Labs' Wristband gives EMG an upgrade

Demo of CTRL-Labs' technology before they were acquired by Facebook.

Founded in 2015 by internet pioneer-turned neuroscientist Thomas Reardon with two more neuroscientists Patrick Kaifosh and Tim Machado, CTRL-Labs was distinct from brain-computer interface startups because the team set out to make a consumer device, a daily-use interface to the computing world. They started to "build devices that allowed them to listen directly to the neurons that are otherwise controlling muscles".

By 2019, when Facebook acquired the startup, CTRL-Labs had developed a wristband prototype that could guide a six-legged robot spider, touch-type, control a wide range of on-screen movements, and more. Since the acquisition, the team has continued to build on its foundational research. Their work on decomposing signals into their constituent motor units, motor unit recruitment, and adaptive learning is taking the EMG field in a very impactful direction. As part of Facebook's 'Reality Labs' division alongside AR/VR, they appear to be leaning hard into interfaces to the virtual world, including a wrist haptic feedback prototype, that allows the user to feel growing resistance while drawing a virtual bowstring.

It appears that Facebook is committed to the wristband tech for the long term, with VP Andrew Bosworth speaking of this technology as an investment with a 5-10-year horizon. Earlier this year, Facebook divested its other BCI projects, such as the implanted speech neuroprosthesis collaboration with UCSF scientists.

A screenshot from Facebook's wrist haptic technology demo.

Cala Health's Cala Trio stimulates nerves at the wrist to impact the brain

The other path-breaking wristband interface — Cala Health's Cala Trio device — has little in common with CTRL-Labs', apart from their investors. Both companies received investment from Lux Capital and GV (Google Ventures). The Cala Trio does not use EMG sensors, is a medical device rather than a consumer gadget, and is already on the market as a prescription device.

Cala Health is the poster child for bioelectronic medicine, a class of therapies that aims to treat chronic diseases by electrically stimulating nerves. Also called 'neuromodulation', this rapidly growing branch of medicine is developing devices that stimulate various nerves to treat cardiovascular, metabolic, or neurological conditions without the side effects that pharmaceuticals often cause.

The primary indication for the device is essential tremor — a kind of arm tremor that affects 7 million patients in the US. In the company's words "Cala Trio gently stimulates the nerves in the wrist to disrupt the tremulous activity in the brain, without the need for invasive brain surgery or medication". Founder and CSO Kate Rosenbluth and CEO Renee Ryan have led the company through a large clinical trial, tracking therapeutic efficacy for months as patients wore the devices at home.

Cala Health's Cala Trio wristband. Screenshot from a company video.

The device is one of the first successful peripheral neuromodulation therapies to come to market. Its presumed mechanism of action is that it stimulates nerves at the wrist to target the ventral intermediate nucleus in the brain. This turns the median and radial nerves into a non-invasive interface to the brain's tremor network. Unlike typical prescription medical devices, the Cala Trio is patient-centered, meaning that it is shipped directly to the patient's home once prescribed, and the patient is in full control of the therapy. The wristband also contains onboard accelerometers that allow it to track the motion of the arm and hand in 3D. This data is returned to Cala Health while the device is charging, and the company can use it to monitor the efficacy of the therapy and improve it further.

The company and recently received FDA's breakthrough designation for Parkinson's disease tremor, signaling an easier path to reimbursement in the US market. Cala Health's last fundraise was a $50MM Series C round in 2019. The team is pushing forward to deliver more neuromodulation therapies for neurological, psychiatric, and cardiovascular conditions — perhaps by stimulating nerves through a wearable device at the ear.

The two devices come from different worlds and have vastly different target users, but both point to the wrist as being the gateway to the brain. Whether it's sensing the intention to move as in the case of Facebook's still-unnamed wristband or stimulating peripheral nerves to affect central nervous system networks in the case of Cala Health, non-invasive access to nerves at the wrist opens up a world of possibilities.

Medical device development is chronically underfunded. As a result, the underlying technology is frequently outdated when it comes to market, and the devices themselves are not always user-friendly. Facebook's investment in accessible, consumer-focused neural interfaces (and the R&D behind them) may well be the catalyst this space needs.

Going Long on the Vagus nerve: Startups Re-discover the Autonomic Neuromodulation Goldmine

Fri, 04 Jun 2021 00:00:00 GMT

Electrical stimulation of the Vagus nerve to treat chronic disease has been around for 35 years, with mixed success. Fifteen ventures, as many therapeutic areas, and billions of dollars later, a new wave of breakthrough devices targeting areas like auto-immune diseases and diabetes is emerging. This post expands on a recent STAT article about companies in the field, looking deeper into the history and further into the future of vagal nerve stimulation devices.

Vagal nerve stimulation (VNS) therapies promise a revolutionary alternative to pharmaceutical treatment for chronic diseases. By electrically stimulating the Vagus nerve, wearable or implanted devices could modulate the activity of the brain and other organs more precisely, conveniently, and with fewer side effects than drugs.

The field has come a long way since the first commercial venture (Cyberonics) was founded in 1987 to develop an implantable vagal nerve stimulator for epilepsy. The Feinstein Institute, DARPA, and the pharmaceutical giant GSK have all made significant R&D investments in bioelectronic medicine. There have been more than a dozen spinoffs from these R&D efforts, but few have had made it past the clinical stage.

As of 2021, only four Vagus nerve devices have made it to market. Patients can only access them in situations where drugs do not work (refractory epilepsy or treatment-resistant depression for example). Payers have been late to the party and have sometimes withdrawn coverage after initially offering it, meaning that health insurance often does not pay for treatment.

This hasn't deterred new entrants, like Nēsos, Spark, or GSK's venture Galvani Bioelectronics, all set up after 2016. They have uncovered lucrative new therapeutic areas and devised more precise ways of stimulating the system. While the early ventures focused on cardiovascular illnesses, depression, and epilepsy, the newer players have their sights set on diabetes or autoimmune diseases like rheumatoid arthritis.

Strictly speaking, not all the companies featured in this post target the Vagus nerve; some target the sympathetic side of the system, or its naturally occurring sensors. CVRx, for example, modulates the autonomic nervous system by stimulating the carotid body. Galvani is working on other autonomic nerves in addition to the Vagus, often choosing to modulate sympathetic nerves closer to their target organ.

The Science

The autonomic nervous system (ANS), which flips the body between the 'fight or flight' and 'rest and digest' states connects to all the major organs and modulates their function through electrical impulses. This twinned system regulates all biological functions: Breathing, heart rate, blood pressure, food intake, metabolism, and even immunity.

The Vagus nerve is the single long nerve that makes up the 'rest and digest' arm and connects with nearly every organ. But it isn't just a straightforward conduit that transmits information top-down from the brain to the organs. It is a bi-directional information superhighway, that also picks up signals from organs and from receptors that monitor the blood for chemical signals and passes them to the brain in a feedback loop. The 'fight or flight' arm of the autonomic nervous system is organized differently, with a chain of nerve nodes along the spinal cord leading to a dedicated nerve for each organ.

This makes the autonomic nervous system a natural target for many chronic diseases. If a device stimulates the appropriate fibers of the Vagus nerve in the correct location, it can selectively influence afferent (towards the brain) or efferent (towards the organs) signaling. A device placed at the neck or the ear, stimulating for a few minutes every day could influence metabolism or immunity to treat a host of conditions.

Vagus nerve and ANS ventures (1987 to 2021)

The New Wave

'New' is relative in the VNS world. Timelines from startup to FDA clearance can easily stretch to a decade or even two.

Spark

Founded in 2018, Spark Biomedical develops wearable neuromodulation devices that treat opioid withdrawal. The device wraps around the ear with a little earplug attachment that allows it to simultaneously stimulate the vagus nerve in the ear and the trigeminal nerve in front of the ear.

In a series of events that can only be described as fortune favoring the brave, the Spark team managed to get the right clinicians on board, find a suitable clinical trial site, secure funding for their first neonatal study, and receive breakthrough device status for this population in record time. They received their first FDA clearance for their use in adults in January 2021 and are now working towards approval to treat opioid withdrawal in newborns. Reimbursement or insurance coverage for their therapy is still a few years away, and the adult device is currently available on a self-pay or charitable use basis.

Although there is no public information about the company's funding aside from an SBIR grant, the team should be able to garner substantial investor interest given their rapid achievement of milestones and the huge unmet need for the product due to the ongoing US opioid crisis.

The idea of stimulating the Vagus nerve at the ear is not new. Cerbomed received approval in Europe (CE marking) for the treatment of epilepsy, depression, and pain through a patient-controlled earpiece-mounted vagal nerve stimulator in 2012, but the company has since fallen off the VNS map.

The Nēsos (left) and Spark (right) devices that stimulate the vagus nerve at the ear. Images from the Nēsos and Spark websites.

Nēsos

Nēsos, founded in 2016, is the brainchild of Nevro founder Konstantinos Alataris. With $16M in funding so far, the company received breakthrough device status from FDA for their earbud-like non-invasive vagal nerve stimulator. The device is used to provide 'e-mmunotherapy' and worn for a few minutes every day.

The company completed a pilot study for rheumatoid arthritis, an auto-immune disease that is currently treated with expensive biologic drugs and is moving on to pivotal trials. Migraine and depressive disorders could be next in the pipeline, along with a new fundraise.

Some of the early team is ex-Nevro, a company that went from founding to IPO in 8 short years, successfully competing with the giants of the spinal cord stimulation industry. A non-invasive device, combined with a commercially attractive launching indication and the founder's track record in the industry will likely take this start-up far.

Galvani Bioelectronics

Galvani is the pharmaceutical major GSK's high-profile bet on bioelectronic medicine. Founded in 2016, through a £540M partnership with Alphabet's Verily, Galvani has a much larger mandate than the typical VNS startup. Their focus appears to be less on the Vagus nerve, and more on the sympathetic arm of the ANS, which allows activation of multiple neuromodulation approaches for diseases like diabetes. The company has not yet announced a launching indication or therapeutic area but is probably well into their pilot human trials, targeting nerves to the spleen to treat inflammatory or auto-immune disease.

Galvani has taken a pharma-style pipeline approach to neuromodulation. They start with investment in R&D, incubating in-house research groups. Approaches that show promise on pre-clinical trials are moved forward to pilot clinical testing and then on to pivotal trials and FDA approval. Galvani has also formed numerous external research collaborations. Their parent GSK has a bioelectronics venture arm called 'Action Potential Ventures'. Verily was originally intended to provide some of the cutting-edge electronics capabilities for the nerve interfaces, but Galvani has also licensed devices from Nuviant medial and Enteromedics.

Other than the recent news about the firing of chairman Moncef Slaoui, Galvani has kept a low profile since its launch in 2016. This company may well be one to watch, given its lineage and funding.

BIOS

BIOS positions itself as a software platform rather than a neuromodulation startup. Founded in 2015, with $5.6M in funding so far, the team has built a machine learning platform off weeks of data collected from vagal nerve cuff electrodes in swine. While working on characterizing this data to discover biomarkers for chronic cardiac or respiratory disease, their larger goal is to provide a read-write interface to the Vagus nerve and the sympathetic chains around the spine so that pharma or other medical devices can discover biomarkers or implement closed-loop neuromodulation. All steps involve a heavy dose of machine learning magic.

BIOS' current team is short on both clinical expertise and industry experience and will likely be taken more seriously once they pass a regulatory milestone or make it to a clinical study.

The Old Guard

Companies founded in the 2000s are now making waves in ANS neuromodulation.

Setpoint

10 years older than Nēsos, Setpoint Medical is also going after the same disease -- rheumatoid arthritis. They also recently (October 2020) received a breakthrough device designation and are now in trials. Setpoint's device is a pill-sized, wireless, remotely rechargeable stimulator implanted near the carotid artery in the neck. The company also has a renowned co-founder in Kevin Tracey. Tracey's research group at the Feinstein Institute pioneered the science of controlling the immune system through the Vagus nerve.

Having raised $216M to date, Setpoint appears well-funded, but the costs and timelines of clinical testing for an implanted device are much higher than a non-invasive device. Once they get approved by FDA, they will still need to convince payers to shell out $30,000 plus the cost of surgery. On balance, this is still far more cost-effective than the expensive biologic drugs currently used to treat the disease. When biosimilars or 'generic' versions of the biologics become available in the US in 2023, this pricing may be harder to justify, especially if the device is only approved for treatment alongside drugs.

They are aiming for multiple inflammatory disorders, an enormous potential market. Setpoint's website indicates that 'Rheumatoid arthritis, inflammatory bowel disease, psoriasis, multiple sclerosis and heart disease all share a common cause: Inflammation.' To realize this value, Setpoint will have to demonstrate that their implanted devices perform significantly better in trials than their non-invasive counterparts.

The Setpoint implantable wireless stimulator (left) and Electrocore's non-invasive Gammacore device (right). Images from the Setpoint Medical and Electrocore websites.

Electrocore

Electrocore, founded in 2005, received FDA approvals in rapid succession from 2017 through 2021 for their non-invasive vagal nerve stimulator to treat and prevent migraine and cluster headache. They have also achieved significant reimbursement milestones in the US and are making headway in the UK, meaning that their therapy is often reimbursed by insurers. The company's website is transparent about its pipeline, which includes several neurological and gastrointestinal disorders.

When applied to the neck for a few minutes each day, the device purportedly stimulates the Vagus nerve at the carotid in an afferent direction — meaning that it influences signals going towards the brain. Skeptics question whether the device affects the Vagus nerve at all, or has an alternative mechanism of action, as clinical trials have not shown that the stimulation produces the side effects commonly associated with implanted VNS devices.

Still, Electrocore is one of only two vagal nerve stimulation companies to have made it this far. The other is Livanova. Both companies are publicly traded, and Electrocore is a fair distance from profitability, with net annual sales at around $3.5M.

MicroTransponder

Founded in 2007, MicroTransponder recently announced the positive clinical trial results for its 'paired VNS' therapy for stroke rehabilitation. The device is an implanted vagal nerve stimulator that is activated during rehabilitation therapy to improve recovery of motor functions.

As part of ongoing clinical trials, the Dallas-based company pairs the same device with headphones for the treatment of Tinnitus. Neither therapy is approved by the FDA yet, and the long interval between development and commercialization means that the design of the cuff electrode with the implantable pulse generator could be outdated by the time the device comes to market.

Microtransponder's implantable VNS system in trials for stroke rehabilitation. Image from the Microtransponder website.

LivaNova

The oldest name in VNS, Livanova, was formed when Cyberonics (founded in 1987) merged with Sorin. Their implantable vagal nerve stimulator received FDA approval for the treatment of refractory epilepsy in 1997 and treatment-resistant depression in 2005. The US Centers for Medicaid and Medicare initially announced that would pay for the therapy in patients but then reversed their decision in a blow to the company and industry. 15 years later, the team hasn't given up on depression, and CMS finally agreed to reimburse the therapy in 2019. They recently kicked off a large (1000 patients, 100 hospitals) study with Verily life sciences for vagal nerve stimulation for treatment-resistant depression. Alphabet's connection is that 300 of these patients in a sub-study will use Verily's wearable device to monitor physiological parameters like sleep, breathing, and heart rate, and a 'Mood app' to record depressive episodes.

Meanwhile, the company's epilepsy neuromodulation business has done well, and the company is expected to be profitable next year, with Wall Street analysts talking of a 'turnaround story'. After a scathing letter from shareholders in October 2020, Livanova will probably move faster to shed the cardiovascular business that it inherited from Sorin, and focus more squarely on neuromodulation.

CVRx

CVRx, another prominent name in ANS neuromodulation, markets a 'Barostimulation' device that does not directly modulate the vagus nerve or any nerve at all. It stimulates baroreceptors in the wall of the carotid artery, which in turn restores autonomic balance, reducing blood pressure and ameliorating the symptoms of systolic heart failure.

Founded in 2001, the company has an interesting backstory. It pivoted from hypertension to cardiac failure in 2006 and then went on to innovate by working with FDA through a unique clinical trial design; finally achieving breakthrough device status, FDA approval, and reimbursement twenty years after it was set up.

PINS Medical

Beijing PINS Medical's VNS device has been approved for the treatment of refractory epilepsy in China sine 2016. The company markets a vagal nerve cuff electrode with an implantable pulse generator. As of 2019, the device was used by nearly 300 patients in 60 hospitals.

The Ghosts

VNS devices have not made it to the market in some key therapeutic areas.

BioControl

Israel-based Biocontrol Medical developed the Cardiofit vagal nerve stimulator to treat cardiac failure. Pre-clinical and early clinical trials were encouraging but the company's 700-patient pivotal clinical trial failed to show that the treatment worked as intended. The company shuttered in 2016, following the failure of its famous INNOVATE-HF trial. Companies like CVRx used the learnings from INNOVATE-HF and Boston Scientific's parallel NECTAR-HF trial to select patients more carefully for their clinical studies.

Transneuronix

Founded in 1995, Transneuronix had developed a 'gastric pacing' device, whose mechanisms of action might have included VNS. The company was acquired by Medtronic in 2005 for USD 260 MM before it completed its pivotal trial in the US. It's not clear if the trial succeeded or if Medtronic is marketing the device or modifying it for other indications.

Enteromedics

Like Transneuronix, Enteromedics had a stomach pacemaker called vBloc, which claimed to treat obesity by blocking conduction in the vagus nerve around the stomach. In 2015 the device received FDA approval for weight reduction, despite mixed performance in clinical trials. IN 2017, a cost-effectiveness study of vBloc therapy for obesity showed that the therapy was likely to be good value for money in certain subsets of individuals. But there was no clear path to national reimbursement coverage and switched to 'prior authorization' to get patients to pay for one-off use.

Recent research casts doubt on the mechanism of action of the vBloc device, indicating that it may work through mechanisms other than VNS. In 2017, the company changed its name to Reshape, pivoting to 'a full-scale provider of medical devices to address the continuum of care for obesity and its associated health conditions' meaning that the company offers multiple obesity-related solutions unrelated to neuromodulation.

Despite the failures, vagal nerve stimulation seems to be picking up pace again, with new research directions and therapeutic areas. Researchers are currently focused on finding ways to stimulate the nerve more precisely, targeting well-defined nerve fiber subtypes. The expansion of VNS into areas lie auto-immune diseases, opioid addiction, and diabetes means that neuromodulation could eventually gain a foothold in some of Big Pharma's biggest markets. If device-makers can convince payers that the technology is cost-effective, this might be the decade when VNS finally takes off.

China's Brain-Computer Interface Landscape in 2021: Has the Dragon Woken up to Neurotech?

Sat, 17 Apr 2021 00:00:00 GMT

Chinese brain-computer interface research is about a decade behind the US or Europe but is likely to catch up quickly as the government pours funding into the China Brain Project. Local manufacturers of implants for Deep Brain Stimulation (DBS) and Spinal Cord Stimulation (SCS) are giving established global players a run for their money. Neurotechnology startup activity is limited.

Global awareness of China's progress in BCI (or any technology) is much lower than the average Chinese person's awareness of Western advances. Chinese media extensively covers news about global BCI events or milestones. Research papers are accessible (and widely accessed) by Chinese researchers, who are part of the international research community. Although Chinese work is starting to be more accessible to a global audience, many Chinese-language publications or media reports are never translated. This post is probably similarly incomplete and perhaps biased, as it is compiled almost entirely from English-language media reports and papers.

Globally developed neural interface devices (say deep brain stimulation devices or cochlear implants) are also widely available in China. China is a critical, fast-growing growing market for these device makers. In addition to these, Chinese physicians have access to locally developed implants and are starting to use them to treat a broad set of illnesses. For example, 61.8% of China's clinical trials of DBS involve non-motor indications (such as neuropsychiatric conditions and opioid addiction), unlike in the West where the primary indications for DBS are Parkinson's disease and essential tremor.

The most convenient way to look at neurotechnology initiatives in China is to look separately at government intent, university research, startups, and established medical device manufacturers — although all these entities are linked.

The Chinese Government and the Brain Project

Brain-computer interfaces, Neurotechnology, and Neuroscience research are all focus areas of the 'China Brain Project', announced in 2016 as part of the 13^th^ five-year plan. The project's ambition is to lead the world in neuroscience, especially applied neuroscience for brain disease within the next few decades. It is motivated by the realization that by 2030 half of the world's people suffering from neurodegenerative disease will be living in China.

Screenshot from the CGTN video about the China Brain Project

This video is a thorough explainer of the Project's goals and the 'One Body Two Wings' vision for brain research. Produced by the state-owned China Global Television Network (CGTN), it allows a glimpse of the brain project as the government intends to portray it, giving a glimpse into the leaders' goals and their perceptions of China's strengths and weaknesses for neurotechnology research. The directors of the project see China as well-positioned to study neurological and neuropsychiatric disease given their large population and emphasize the need for better characterization of these illnesses in the Chinese population. In addition to neurological disease research, they also have an eye on medical devices and neuromodulation. They concede that the US will continue to lead the world in basic neuroscience research for many years to come.

A key feature of the Brain Project is the International Primate Research Center in Shanghai, a facility that the leaders of the project hope will spur international collaboration. They plan to use the facility to conduct functional research and to develop new primate models of neurological disease.

Universities and Research Institutes involved in Neurotechnology and BCI

Money invested by the government goes to dedicated brain research institutes or universities across the country. These centers recruit research groups and investigators, often with a focus on attracting talented Chinese(largely) individuals that have been trained overseas. Some of the top centers in China for BCI research are the Tsinghua University, Tianjin University, Zhejiang University, the newly established Institute of Neuroscience in Shanghai, and the Chinese Institute for brain research, Beijing.

Zhejiang University: Developed China's first brain-computer interface implanted in a human in 2020, announcing a successful implant in a paralyzed individual that allowed him to manipulate robotic arms with his thoughts to shake hands, carry drinks, eat fried dough sticks and play Mahjong.

Image from China News Service

Chinese Institute for Brain Research, Beijing: Launched 2018, the Beijing Municipal government is funding the institute with with USD 29MM in the first year, with a future USD 65MM annual budget.

Institute for Neuroscience, Shanghai: Part of a Neuroscience research science park being built in Shanghai and funded by the Shanghai Municipal government, one of the highlights of the institute will be the International Primate Research Center, to house as many as 6000 non-human primates. Positioned as a world leader, the center has already attracted international scientists.

Tsinghua University: The National Engineering Laboratory for Neuromodulation, Tsinghua University was responsible for the first locally developed DBS device, now commercialized by PINS Medical, Beijing, and used widely within China.

Tianjin university: Source of innovations such as a 'Brain Talker' computer chip designed for BCI applications, an exoskeleton rehabilitation hand robot, and a 'Mind-Typing cap' that allows the user to brain-type 69 Chinese characters per minute, exceeding the speed of using touch screen mobile phones to type.

Clearly, in a country as large as China, there are many more than these 5 top-tier universities working on neurotechnology, including in areas such as materials science, electronics, and neuromorphic computing.

Startups

Startup activity in brain-computer interfaces or neurotechnology appears limited, with sporadic funding announcements, and not many demos or products available in the English-language media. The startup Neuramatrix incubated by Tsinghua university — generated a bit of a ripple earlier this year when they made a funding announcement. Founded in 2019, the company employs about 50 people, mainly R&D and product engineers, and claims that their invasive BCI chip rivals Neuralink.

Screenshots from the website of a company named 'Naolu' showing their 'Brainup' product (left) whose marketing graphics and headband appear to be 'inspired' by Muse (right).

This list of Chinese BCI startups is largely comprised of developers of non-invasive headband-type devices for meditation, biofeedback, and other consumer applications.

Medical Devices, Implants and Neuromodulation

While university research and startups are still nascent, Chinese manufacturers of DBS and other neuromodulation devices have advanced quite far in the last decade. DBS in China is used not only for Parkinson's disease but also for non-motor diseases. Ruijin Hospital in Shanghai, for example, runs a DBS research center for addiction, Tourette syndrome, depression, anorexia, and other neuropsychiatric conditions.

The Chinese manufacturers PINS Medical and Sceneray offer DBS systems on par with, if not more advanced than their Western equivalents. Both companies' devices are approved for use by the Chinese FDA, have the European CE mark and have been introduced to markets like Pakistan, Bangladesh, and Indonesia. Both companies offer advanced remote wireless control of the implants, with well-established methods for remote programming and calibration of the devices and monitoring battery levels through telemedicine by clinicians that may be far away — a feature that has been rendered invaluable during the COVID-19 pandemic. Additionally, both companies are expanding to invasive and non-invasive vagal nerve stimulators and spinal cord stimulation devices.

Screenshot from Sceneray's video showing their remote DBS device reprogramming feature

China is also the fastest-growing market for cochlear implants, with the Hangzhou-based company Nurotron gaining market share from Australia's Cochlear, Swiss-based Sonova, and Austria's MED-EL not only in China but also in other emerging markets.

It is yet to be seen how much original neurotechnology research will come out of China, but it is already clear that the country will be key to the commoditization of these interfaces. In turn, this will facilitate access to this technology globally and make them more affordable for

How DARPA drives Brain Machine Interface Research

Sun, 22 Nov 2020 00:00:00 GMT

The US Defense Advanced Research Projects Agency invests millions in brain-computer interface projects every year, effectively driving the BCI research agenda. This post catalogs the various programs and their recipients, tracking DARPA's investments over the decades

Nobody has funded the BCI research and development with as much sustained energy as DARPA. Almost every advance or major technology in the field can be traced back to DARPA funding to the researchers. Neuralink's 'sewing machine' surgical robot can be traced to the 5-year $70M SUBNETs program, the initial concept for the Stentrode was funded by the RE-NET program, and the prosthetic limbs used by early Braingate trials were funded by the 'Revolutionizing Prosthetics' program. Startups like Nia Therapeutics, Paradromics, and the recently acquired Iota Biosciences also owe DARPA either for direct investment or for funding of the underlying technology.

The Agency funds multi-million-dollar research programs that typically last 4 years and have very specific ambitious goals. DARPA announces programs publicly, and candidates compete individually or by forming consortia to develop competitive proposals. Successful awardees or recipients are called 'performers'. The research that is funded usually shapes the future of the entire field for a few years.

The overarching goals of DARPA research funding for BCI are

Augmenting the cognitive capabilities of its US armed forces personnel
Mitigating the human consequences of armed conflict by developing solutions like prosthetic limbs or PTSD therapies.

Almost all programs fall into one of these two categories, but DARPA has also funded a considerable amount of basic neurotechnology research to meet these goals. The military may or may not use the results, but the technology often finds its way into the general public sphere, benefitting the global neuroscience community. Occasionally, investors, corporates, or billionaires pick up promising tech to commercialize it further. Neuralink and Stentrode are on their way to human trials, working with the FDA, but not all DARPA-funded projects succeed. Kernel's early efforts with invasive prosthetics that stemmed from the DARPA-funded REMIND program, were quickly dropped in favor of more consumer-friendly wearables.

Sometimes, DARPA drops programs midway or changes their names, leaving little trace. The 'Accelerated learning' program on tDCS is an example, with press reports and presentations explaining the goals, only for the program to drop off the DARPA website entirely. Conspiracy theories about DARPA and BCI research abound, but most of these information gaps can be chalked up to changes in policy or strategic direction, or just research efforts that didn't pan out.

DARPA BCI Programs through the decades

DARPA's strategic interest in BCI started in the 1960s, when JCR Licklider, the first director of the Information Processing Techniques Office (IPTO) published his vision of 'man-computer symbiosis'. The first batch of funding was in 1974, under a program called Close-Coupled Man/Machine Systems, later renamed Biocybernetics. The Biocybernetics program is documented in places with forbidding names like 'The Black Vault', but it's not clear at all that there were any sinister intentions behind it. The documents on Biocybernetics are a great read, explaining how DARPA funded universities for research on EEG, EMG, pupillometry, event-related potentials, MEG, and related technology. They also are surprisingly candid, admitting to burocatic missteps, funding lapses, and a 'checkered history reflecting programmatic, organizational and personnel changes at DARPA'. The Biocybernetics program was the first consequential BCI funding in an era when all brain science was 'psychophysiolocial', and most notably led to the discovery of event-related potentials and the P300 component.

The 1980s and 90s are devoid of any documented DARPA investment in this technology, but DARPA's interest picked up again in 2002 and has since grown considerably. DARPA funded both Invasive and non-invasive BCI research in the 2000s, under the umbrella of the AugCog and HAND programs. The HAND program was especially broad in its goals, with funding stretching from 2002 well into 2015.

In the late 2000s, DARPA started to expand and split up this funding into multiple, more specific programs. The Revolutionizing Prosthetics was one of the biggest and most successful, leading to the development of modern functional prosthetics. DARPA also invested heavily in memory enhancement and recovery through the RAM, RAM-Replay, and REMIND programs. Around the same time, another set of programs — Silent Talk, Accelerated Learning, NIA, CT2WS — were all funded to improve performance at various cognitive tasks, and communicating non-invasively with machines. Another program, RE-NET, funded basic and translational research to solve the problem of long-term stability of neural interfaces. The current decade has seen more focus on peripheral interfaces, with the Electrx, HAPTIX, and TNT programs.

While each new program manager redefines priorities, DARPA seems to learn from any failures and adapt accordingly. New programs have started focusing on cooperation with FDA early in the R&D process, and calls for proposals now routinely emphasize ethical, legal, and social aspects of the technology. New programs have clear goals for paths to human use — for both invasive and non-invasive technology. Having observed that it will take decades before invasive brain implants can be used in healthy individuals, DARPA is now also focusing on investments in non-invasive or 'minutely invasive' technology for these applications.

The People Behind the Programs — DARPA Program Managers

Each DARPA program is the brainchild of a program manager. These are powerful and interesting figures, handpicked for their role, which is to set ambitious goals for BCI technology and to shape its research trajectory. They have a limited tenure at DARPA- typically 4 years, but most make their mark on the field during that time. One of the best ways to learn about DARPA programs is to listen to DARPA program managers, where they explain how they got the job, why they picked the priority areas they did, and learn how DARPA works. DARPA hosts a podcast series, where program manager Eric Van Gieson has been featured speaking about some current programs. The Neural Implant Podcast has also hosted 40-minute interviews with DARPA program managers Jack Judy, the architect of the RE-NET program, and Doug Weber, who speaks about his enthusiasm for peripheral interfaces.

Who gets DARPA funding?

Over the decades, the profile of DARPA BCI research 'performers' has also evolved. In the early 2000s, most BCI contracts went to large corporations — think Lockheed Martin, Boeing, and Booz Allen Hamilton — but the newer programs are more focused on universities. In recent years DARPA has funded several small businesses or startups, most notably Paradromics in 2017. Performers are almost always are from the United States, with a few exceptions, such as the Stentrode funding to the University of Melbourne.

DARPA contracts are large and many recipients are large consortia. Universities and corporations often partner to receive funds and sub-contract them. For example, the Revolutionizing Prosthetics program was led by two teams of investigators, DEKA and The Johns Hopkins University Applied Physics Laboratory (JHU/APL). DEKA worked with two universities and a private developer, while APl subcontracted to nineteen primary level contractors (universities and private companies) with ten second-tier subcontractors and collaborators from six countries overall. Of late the number of recipients is smaller, even for large grants.

DARPA also funds BCI research outside these named programs, through mechanisms like open calls and small business innovation grants.

Conclusion

There is no other agency in the world that funds neurotechnology research with the unmitigated focus that DARPA does. It's no wonder that their funding has driven the research agenda for neural interfaces for decades. It's anybody's guess what DARPA's next BCI program will fund, but the agency is certain to continue to steer the BCI community's research efforts for many years to come.

On Neural Dust

Sun, 25 Oct 2020 00:00:00 GMT

This post reviews the significant milestones in the development of neural dust and lays out the state of the art in 2020.

The neural dust device is 3 millimeters long and 1×1 millimeters in cross section. Picture: Ryan Neely, UC Berkeley

The Sci-Fi Noir video series 'DUST' imagines the dystopic future of brain-computer interfaces. Science fiction novels are replete with the idea of 'Neural Dust', or nanoparticles that float around in the brain to sense its activity in real-time. The phrase 'neural dust' suggests a state where countless untethered microscopic specks are sprinkled among neurons, collecting information to put a bigger picture together.

As with most brain-computer interfaces, the technology that we currently have is far behind the fiction. The goal that neural dust researchers are currently pursuing — 10-100um free-floating sensors that detect and report local extracellular electrophysiological data — is relatively modest compared to the grand idea that the name evokes.

Scientists have indeed developed micron-scale devices that can read and write from neurons. Some researchers have even gone as far as in vivo testing. But overall, this category of BCIs has received much less attention than the fixed implantable devices or the wearable headgear. In a recent podcast (Jump to minute 50 of the to hear the discussion on neural dust), the CEO of Paradromics dismissed the idea, suggesting that the delivery barriers are insurmountable: "I'm not sure neural dust makes sense as a concept. You're either delivering it like a gene gun — ballistically — or you're sprinkling it on top, in which case it's no different from an ECoG grid."

The following week, the pharma giant Astellas announced that they were acquiring neural dust startup Iota Biosciences for hundreds of millions of dollars, doubling down on their earlier investment in the company.

Iota Biosciences, whose tagline is 'Vanishingly Small Bioelectronics', holds an exclusive license to the millimeter-sized bio-electronic device technology developed at the University of California, Berkeley. Michel Maharbiz, and Jose Carmena, who developed the technology through DARPA-funded research, founded the company in 2017. Rather than build a single 'BCI' or neural interface, the company's goal is to develop a customizable platform for research or treatment of disease.

The researchers coined the term and explained it in a 2013 paper, and followed it up by developing a prototype designed to be free-floating and stay on the surface of the brain. The group also validated that the technology worked in vivo. Their technology consists of a millimeter-scale sensor coupled with an ultrasonic backscatter system for powering and communicating with it, demonstrated to work with spinal nerves rather than in the brain. At 1mm, the implant isn't yet 'vanishingly small', but the use of ultrasound to power and communicate with the device was a game-changer.

The neural dust device tethered to a spinal nerve. Picture: Ryan Neely, UC Berkeley

Interestingly, the first author on all these papers, DJ Seo, is a founding member of Neuralink, now heads implant systems at Neuralink. Neuralink, of course, is not working on neural dust but on a coin-sized implantable device, electrode threads, and a robot sowing machine.

In 2016, Rikki Muller and her research group developed an adaptation of the neural dust motes called 'StimDust)' that could also stimulate nerve fibers. They took the neural dust platform and built a stimulator that can wrap around a nerve cuff to stimulate it while also recording and transmitting the data. The key innovation was a custom integrated circuit to transferred ultrasound charge to the nerve in a well-controlled, safe, and efficient way.

Meanwhile, Arto Nurmikko at Brown University and Vincent Leung at the Qualcomm Institute Circuits Lab at UC San Diego are developing what they call "Neurograins". Neurograins are fully wireless microscale implants that may be deployed to form a large-scale network of untethered, distributed, bidirectional neural interfacing nodes capable of active neural recording and electrical microstimulation. They aim to create a "cortical intranet" of tens of thousands of wireless micro-devices — each about the size of a grain of table salt — that can be safely implanted onto or into the cerebral cortex, the outer layer of the brain. The implants will operate independently, interfacing with the brain at the level of a single neuron. Charging and communication will be carried out through radiofrequency waves. The effort was funded through DARPA's Neural Engineering System Design (NESD) program. An engineering scheme appears to be in place, but it isn't clear that the researchers have tested the system extensively in vivo.

In 2019, a team of researchers at Stanford, led by Guosong Hong developed a technique they called 'sono-optogenetics'. Although they didn't use the term neural dust, their approach, explained in a paper called 'Seeing the sound' comes closer to a neural dust vision than anything we have seen so far. The method they developed uses circulating light-emitting nanoparticles to activate the motor cortex in mice whose neurons are genetically modified to contain light-sensitive receptors. These nanoparticles are 'mechanoluminiscent', meaning that they transiently light up to activate the surrounding neurons. A non-invasive ultrasound beam controls the nanoparticles, turning them on and off with sub-millisecond precision. The 700um focus of this beam can be anywhere, including deep within the brain. Changing the focus of the ultrasound activates different neurons, provided that they have light-sensitive photoreceptors. The other trick that the paper describes is that the particles are rechargeable, or rather that they need recharging. The charging occurs through an external light source when the particles pass through blood vessels close to the surface of the skin. The highly innovative system showed promising results in vivo and is a testament to interdisciplinary knowledge. However, it is not likely that optogenetics will be in humans in the near-term outside some select applications for the retina.

The sono-optogenetics system tested in rodents. Picture from Seeing the sound , Guosong Hong, 2020

The Astellas acquisition of Iota is promising and means that the electroceutical technology is likely to enter human trials and move towards regulatory clearance. But it's not clear that there are any startups or even academic research labs still working on the neural dust approach. The handful of technologies described above appear to have just touched the surface of the concept, with many barriers that exist before the technology is widely applicable. Size, scalability, and most important — delivery mechanisms — appear unsolved, at least until optogenetic techniques become widely feasible in humans.

Closed-loop brain-computer interfaces

Sun, 27 Sep 2020 00:00:00 GMT

All BCI devices record. Only a few of them provide direct feedback to the brain in real-time. This post lists current closed-loop human BCI devices and their applications.

X-rays from patients implanted with the Neuropace RNS device for epilepsy treatment. Picture from Morell et al. 2014

In a roundabout way, all our computers and phones are closed-loop brain-computer interfaces. Let's narrow the focus to BCIs that communicate 'directly' with the brain, bypassing our natural mechanisms for input (vision, hearing, touch, taste) or output (speech, writing/typing, or gestures).

What qualifies as a closed-loop brain-computer interface (BCI)? A synthetic interface with the brain that records neural activity and processes it in real time to decide when to stimulate the brain directly or indirectly.

A system that reduces epileptic seizures by recognizing emerging patterns in seizure foci in the brain and then responding to quell them is a closed-loop BCI. In contrast, current Deep brain stimulation techniques used in the treatment of Parkinson's disease are not a closed-loop BCI. Their stimulation pattern is fixed, and the electrodes don't record activity or respond to it.

Does a closed-loop BCI have to directly stimulate the brain through electrodes? Not necessarily. EEG headbands that provide biofeedback for meditation also qualify as closed-loop BCIs when they provide auditory or visual feedback that responds in real time to brain activity.

A closed-loop BCI by definition is a device that reads from the brain, rather than other parts of the nervous system. Biofeedback from heart rate or breathing rate, reading from EMG or other peripheral nerves, therefore, does not qualify as a closed-loop BCI.

Some of the best-funded players in BCI don't seem to be reaching for closed-loop BCI, at least initially. Neuralink, Paradromics, Kernel, and Neurable have all focused their prototypes and papers on recording from the brain and interpreting the information rather than on closing the loop.

But several players do have closed-loop BCI as their goal. They aim to augment meditation, sleep, memory, and even to treat epilepsy. This post lists current human closed-loop BCIs, with a focus on devices that are or should soon be commercially available.

##Meditation

The MUSE headband is the classic example of a closed-loop meditation BCI application. The 'Brain Sensing headband that Improves your Meditation Practice' translates brainwaves into soundscapes that allow the user to monitor their brain activity while meditating. If the user hears stormy sounds, they can calm their brain rhythms until they start to hear more peaceful sounds and 'reward chirps'. The MUSE uses 4 EEG sensors incorporated into a headband that goes across the forehead and behind the ears.

As far as non-invasive BCI goes, the Muse seems to be the perfect example of a product that is useful to people right now. The headband has mixed reviews, with some users complaining that the effectiveness of the feedback depends heavily on your state during the calibration session. The newer version seems to rely a lot more on other sensors (Photoplethysmography (PPG) sensor, pulse oximeter, accelerometer, and gyroscope) than the EEG to provide biofeedback.

Other comparable products like the lower-priced 'Flowtime' are also available for purchase.

Muses's meditation and sleep headband. Picture from Gadgets and Wearables review

##Sleep

While the MUSE offers some sleep features, the Dreem 2, billed as your 'personal sleep assistant' makes a sleep profile for viewing in the morning and coaches the user on how to sleep better. The 6-electrode headband is tailored towards long-term sleep monitoring rather than any real-time feedback that enhances sleep.

The audio on the headset, delivered through bone conduction, automatically lowers and switches off if the user's EEG signals indicate that they have fallen asleep. The EEG-based sleep detection capability also determines when to set an alarm to wake the user. With these 2 features, the Dreem 2 just about makes it into closed-loop BCI territory.

##Attention

There does not appear to be a device that focuses on enhancing attention through BCI. The closest device is the Emotiv, whose website says 'The EMOTIV MN8 helps to measure and analyze the changes in your employees' levels of stress and attention using EEG and EMOTIV's proprietary machine learning algorithms'. The app that accompanies the device presumably displays stress and attention levels across the day or during activities. It is not clear how (or if) the feedback is used to guide or nudge workers.

Emotiv's representation of the use of a BCI device for monitoring stress and attention at work

The gaming industry appears to have invested in research into closed-loop BCIs that respond to attention-like cues gathered directly from EEG. Valve's experimental psychologist describes using signals from BCI to engage the player for longer, responding to signs of stress, tiredness, or boredom by changing the level of difficulty of the game. Smaller developers like Brainattach have also experimented in this direction.

##Memory

In 2018 a closed-loop invasive device showed in human trials that hippocampal recording and stimulation could boost recall. The 'hippocampal memory prosthetic' facilitates encoding using the subject's own hippocampal spatiotemporal neural codes for memory. When tested across 8 subjects, there was significant improvement (35%) in both short-term and long-term retention of visual information. The technology was picked up by Kernel (with its developers), and the research was furthered there, but the prosthetic was since shelved. Kernel has moved into non-invasive neural interfaces that read from the brain and decode signals, with no stimulating capabilities.

Meanwhile, Nia Therapeutics is developing and commercializing an ECoG-based 'memory augmenting' closed-loop system that senses from the temporal lobe. The device uses electrodes that record the state of the subject's brain, analyze it, and decide whether to trigger stimulation. A study of a prototype version showed a small improvement in subjects' ability to remember words, using a machine learning algorithm to close the loop. Now, coupled with a Cortera Technologies' wireless neuromodulation device, Nia Therapeutics is further developing the prototype, focusing first on a device targeted at patients with traumatic brain injury.

##Epilepsy

Neuropace's neurostimulation system for epilepsy treatment is the only invasive FDA-cleared closed-loop BCI. Depending on the location of the epileptic focus in the brain, a surgeon places deep and/or surface leads that sense patterns that are likely to lead to a seizure, and automatically deliver pre-programmed stimulation pulses to prevent it. Their study, completed in 2020 showed that 30% of patients achieved more than 90% seizure reduction in the most recent 3 months. There are no numbers available on how many patients are currently using the device, but the company recently received a second FDA approval and raised funding to support commercial expansion.

Neuropace's RNS system

The experimental invasive hippocampal and temporal lobe BCIs seem very far from the everyday usability of the sleep-monitoring and meditation-enhancing headbands. What they all have in common is that they interpret the language of our neurons and allow us new ways to talk back to them — without having to learn their language first.

Funding for Brain-Computer Interface Ventures

Fri, 28 Aug 2020 00:00:00 GMT

Can following the money tell us about the interface between academic research and venture-funded companies in BCI. Is there a clear boundary at all? What does the 'funding lens' reveal, about who is moving the field forward? How far along are the new BCI companies compared with the incumbent neural interfaces?

Funding is a measure of potential, and occasionally a predictor of success. Regardless, tracking investment in brain-computer interfaces is a useful way to assess how the new generation of BCI developers stacks up. Interest in BCI has been growing, and we're likely to see a new wave of investments in 2020 and beyond. Here is the picture in August 2020, ahead of Neuralink's second big announcement.

At first glance, it seems like European companies get funded to do invasive medical applications and US companies work on consumer applications with the notable exceptions of Paradromics and Neuralink. This is probably not true. Many US research efforts towards invasive BCI — like 'Braingate', perhaps the most advanced BCI in human trials today — are structured as academic groups. Across the Atlantic, companies tend to spin off or license out technology developed in universities before human trials.

Companies in the Bay Area are highlighted as a separate group, because of the sheer number of BCI ventures headquartered there, and because these ventures have different levels of access to funding compared to those elsewhere in the US. However, that label may not be fully accurate. For example, Paradromics was started up in Silicon Valley and subsequently moved to Austin, Texas. Similarly, Emotiv started up in Australia and subsequently set up headquarters in San Francisco.

'Kernel' stands out in its own 'Research Devices' category. The company provides 'Neuroscience as a Service (NaaS) gives you on-demand access to our world-leading brain recording technology.', allowing clients to perform non-invasive neuroscience experiments remotely through it's 'Flow' and 'Flux' platforms.

But Kernel is not alone in catering to the research community. 'Blackrock, Ripple, TDT, and Plexon all produce products, components, and services for BCI research. These include lab electrophysiology equipment, FDA-cleared microelectrode arrays for human implantation, custom BCI development services, human BCI implanted arrays, including consulting services, and technology for optogenetic stimulation. These are privately owned, but with little information on funding available many seemed to have originated as spinoffs from university research groups. It seems that following the money does not always reveal where new technology is being developed. 'Neuropixels', a publicly funded research and engineering effort that developed a popular silicon CMOS digital neural probe, also deserves a mention in the 'research devices' category.

Finally, and perhaps most importantly, neuroscience researchers will attest to the contribution of community-developed efforts like OpenBCI, who develop widely used open-source non-invasive BCI tools and Open Ephys, an organization that consolidates open-source software efforts into tools that are used by the likes of Neuralink and advocates for common open standards that move the field forward. These organizations are fundamental to moving BCI research forward, but not visible at all when you use the 'funding' lens to look at the field.

Even companies like Neuralink who purport to want to eventually build a consumer device, are building medical devices first. Let's narrow the focus to invasive neural interfaces for medical use and include some of the older neural interfaces that were conceptualized in the 1990s and early 2000s.

Technology that requires surgery to implant an interface that will stay in the body for years, stimulating or recording from the nervous system throughout its lifetime requires approval from regulators. This involves years of pre-clinical testing and clinical trials to demonstrate that the implants and the surgical procedure are safe. Developers of invasive neural interfaces follow a different path from developers of consumer devices.

The technology that the newer ventures like Paradromics are building may be light years ahead of the large sensing leads and relatively imprecise neural stimulators that are currently approved by the FDA for the treatment of epilepsy, Parkinson's disease, heart failure or chronic pain. Their paths to commercialization, however, will depend strongly on the outcomes of their clinical trials. In this sense, they are closer to these incumbent medical device manufacturers, and to developers of motor, sensory, and visual neuroprosthetics. Hopefully, the newer generation of BCI devices will have shorter timelines before clinical use, with FDA announcing its support for BCI devices and putting out guidelines for their trials.

To shed more light on how these technologies are brought to life, it's useful to dig deeper into where the technology is developed and commercialized. All technology builds on previous discoveries. Some inventions are explicitly transferred from the research world to the commercial sphere through 'technology transfer agreements' or licenses. It's not unusual for the initial R&D to be done through a research consortium and handed over to a commercial entity who is responsible for bringing it to market. This can also be done through a 'spin-off' from a university.

For example, the Australian Federal Government awarded a $42 million grant to a research consortium called Bionic Vision Australia to develop bionic vision technology. The technology was later transferred to Bionic Vision Technologies, who went on to raise $18 million from private funders, presumably to further clinical trials and further development.

In the case of BCI, it appears that the 'wearables' or non-invasive consumer BCI products are mostly built in-house at tech companies. The medical neural interfaces or the 'implantables' have are more likely to be developed in a university lab and transferred or spun out for commercialization after a certain degree of success has been proven.

Public institutions are closely linked with neural interface research and development globally. More than 60% of the 36 neural interface companies either build directly on technology licensed from a university or receive public funding in the form of grants for research and clinical trials. The US National Institutes of Health (NIH), the US Small Business Innovation Research (SBIR), the European Commission's Executive Agency for Small and Medium-sized Enterprises (EASME) and the European FP-7 R&D programs have all been significant funders of BCI and continue to award grants to further the technology.

Another notable agency involved deeply in United States BCI development is the Defense Advanced Research Projects Agency or DARPA. More than half of invasive neural interface technology companies in the US are directly or indirectly funded by DARPA. The agency has taken an interest in BCI development for decades and continues to fund its development.

With the 'Elon Musk spotlight' that is currently on the field, brain-computer interfaces are likely to see an uptick in investor and researcher interest over the next few years. Hopefully, this means that new safe and more effective technology will come to market soon.

####Exclusions and Omissions

This post does not include companies

For whom funding information is not publicly available (Neurosity, Neurosky, Melomind, Neuro Device, MED-EL, Advanced Bionics — acquired by Sonova, Advanced Neuromodulation Systems — acquired by St Jude, Marsi Bionics and Bionic Sight)
Large medical device manufacturers, where neural interfaces are not the sole business (Medtronic, St Jude, Sonova)
Companies that are no longer in business (Retinal Implants AG)
Wetware companies (Koniku, NETRI)
All EEG headset manufacturers (too many to list)

####Edits

31 Aug 2020: Added Neuropace funding USD 67M to chart 2

Neural interfaces for vision: Is there light at the end of the tunnel?

Tue, 25 Aug 2020 00:00:00 GMT

Current retinal implants deliver just enough vision to enable the blind to see bright lights, high-contrast edges, and moving objects. Considering their enormous potential, visual prosthetics have lagged behind other BCIs, because we still don't fully understand what happens between the retina and the cortex. Will the new generation of visual interface technology now in trials — optogenetics, cortical implants, and 'retinal coding' — finally allow the blind to see better?

Recipients of current visual cortical implants see little more than points or bars of light. Photo by Asim Z Kodappana

Vision is by far the highest-bandwidth and fastest information input system into the human brain. The bandwidth of the human retina is approximately 8960 kilobits per second. The visual system aggregates this information at multiple levels before it is perceived, acted on, stored, or forgotten. The first level of complex processing occurs at the retina, where 100 million photoreceptors perceive light and pass on a high-fidelity representation to the optic nerve. Since the optic nerve has only 1 million fibers, it is evident that information from the retina is resampled and compressed both spatially and temporally by the local retinal neuronal network.

Our understanding of the retinal encoding process is incomplete at best. The encoding of visual information grows even more complex along the relay from the optic nerve to the visual cortex. This is why most attempts at visual neuroprosthetics aim to activate cells as close as possible to the start of the system — at the retina.

In 2020, there is a single retinal implant approved by the United States FDA, and two approved by the European regulator. These implants allow people to detect light sources such as airways and lamps, follow high-contrast edges, recognize moving objects. Some implant recipients can recognize large letters at close range, but this level of visual acuity falls far short of the 'legally blind' threshold.

It is precisely because the visual system is so advanced that we haven't yet been able to mimic it. Researchers have tried neural stimulation at various levels of the visual pathway

#Neuroprosthetics and the visual pathway

###Retina

Most visual neural interfaces operate at the level of the retina, bypassing photoreceptors that are damaged in certain hereditary or age-related forms of blindness. These require that the circuitry downstream of the retina up to the visual cortex be fully developed and intact. Since much of the visual circuitry in the brain is developed in infancy, in conjunction with a seeing eye, these implants apply to forms of blindness that are acquired in adulthood.

At least 6 retinal implant designs have passed preclinical testing, and are currently or will soon be in clinical testing. Depending on their design, these implants are placed within the layers of the retina or just outside it. They typically consist of a flat microelectrode array (16-60 electrodes) that simulates the bipolar and/or ganglion cells of the retina, bypassing the degenerated photoreceptor layer. Except for the IMS Alpha, which uses internal, light-sensitive photodiodes, retinal implants consists of a camera mounted on eyeglasses and a visual processing unit. The external components capture and process images, then activate the implanted microelectrodes, which in turn stimulate the retina. This design allows developers to program the pattern of stimulation and potentially improve implant function through software upgrades.

Second Sight's Argus II is the only implant approved for use by the FDA and has over 350 global users. The 2013 approval by FDA was a humanitarian device exemption that provides a 'reasonable assurance of safety and probable benefit' but stops short of supporting any claims that the device is effective. In the manufacturer's words, 'The Argus II is authorized by Federal (U.S.) law to provide electrical stimulation of the retina to induce visual perception in blind patients with severe to profound retinitis pigmentosa and bare light or no light perception in both eyes. The effectiveness of this device for this use has not been demonstrated.'

This is a far cry from the language used in the FDA approvals of cochlear implants (neural interfaces for hearing), which state unambiguously that these devices improve speech understanding in noise, sound localization, and self-perceived quality of hearing.

The Argus II is also approved for marketing by the European regulators, as are the IMS Alpha (2013) and Pixium vision's IRIS II (2016).

Other retinal implants include Bionic Vision Technologies' Bionic Vision implant which has been tested in humans, and the Boston Retinal implant, whose developers plan to develop a chip with at least 200 electrodes before human testing.

###Optic Nerve

Interfaces that directly stimulate the optic nerve broaden the scope of visual neuroprosthetics to include those with total degeneration of the retinal neurons. Early stimulation studies of the optic nerve using 'cuff' electrodes evoked visual sensations broadly distributed across the visual field. There is also evidence that optic nerve fibers respond better to a 'train of stimuli', suggesting that they are able to integrate information over time and space. More recent pre-clinical studies on intra-neural stimulation showed that each stimulating electrode introduced a specific pattern of cortical activation.

However, there are no current optic nerve stimulatory devices in clinical trials.

###Cortex

The visual cortex has also been a target location for interfaces, beginning in the 2000s with the controversial Dobelle eye.

The CORTIVIS system, developed at the Universidad Miguel Hernandez de Elche is a visual cortical BCI currently in clinical trials. The prosthetic is composed of a modified 100-electrode Utah array, coupled with custom software that matches simple visual inputs with retinal outputs, to tune each electrode independently.

Second Sight's visual cortical prosthesis, 'Orion' is also in clinical trials. The 60-electrode device is stimulated by processed signals derived from a camera mounted on a pair of eyeglasses to capture images. The key to this technology appears to be 'a technique called dynamic stimulation, where the device stimulates the brain in sweeping patterns across the array of implanted electrode'.

For now, these cortical implants produce little more than points of light. Clinical testing appears to be directed at translating images into patterns that are meaningful to the cortex — an uphill task considering the complexity and depth of the visual system.

There are more cortical implants in pre-clinical stages, notably a 1000-electrode implant developed at the Netherlands Institute for Neuroscience, and a multi-array approach with multiple simultaneously implanted 16-electrode arrays under development at the Illinois Institute of Technology, Chicago.

###Optogenetic Techniques

Optogenetics is also making headway in visual prosthetics, with at least 3 clinical systems in development.

Bionic Sight's technology is based on an algorithm that reconstructs signals that the retina would generate when exposed to an image or a video and transmits these directly to the optogenetically modified ganglion cells whose axons make up the optic nerve. Pre-clinical studies show that these algorithms are able to encode images in much the same way the retina does, providing output that the rest of the visual pathway should be able to interpret. Human safety and feasibility trials started in 2020, with increasing doses of the gene therapy vectors that target optic nerve cells to make them light-sensitive, so they can subsequently be stimulated with the 'retinal code'.

Paris-based Gensight Biologics also has a safety/dose clinical trial ongoing to target retinal ganglion cells with a light-sensitive protein, rendering them photoreceptive. This method will be coupled with biomimetic googles to capture, amplify, and present incident light at the appropriate wavelength.

Researchers at the SUNY Downstate Medical Center are developing an intra-cortical optogenetic therapy system called OBServ. The system consists of consisting of optogenetically modified visual cortex neurons that act as photoreceptor cells, coupled with two external cameras and extensive processing of the visual signals.

#Challenges facing Visual prosthetics

Despite the multitude of research groups working on the problem — there is still a long road ahead with visual prosthetics.

It was 20 years from conception to the first FDA approval of a retinal implant. A decade later, despite the sizeable number of candidates, no other visual prosthetic has met this bar. What makes it so hard to develop a visual interface that works?

First, the approval process is lengthy. The degree of invasiveness of retinal or cortical implants necessitates extensive pre-clinical and clinical testing. The Argus II went through lab studies of the implant, bench testing, preclinical studies of the design, animal testing for safety, and a 30-subject safety trial before it was approved for humanitarian use.

Second — somewhat surprising for technology with such life-altering potential — funding doesn't seem to stretch to clinical trials. Second Sight's last round of funding in October 2019 was a USD 2.4 M grant from the NIH, for clinical trials of the Orion. In March 2020, the company announced layoffs of 80% of its workforce and intent to wind down in response to the impact of the global COVID-19 pandemic on its ability to secure financing.

For some implants, the designs that seemed to work in pre-clinical tests simply do not show the desired results in clinical trials. The manufacturers of the CE marked Alpha IMS II discontinued business activities in 2019, citing the innovation-hostile climate of Europe's rigid regulatory and unsatisfactory results in patients.

Perhaps the biggest challenges facing the field relate to the underlying neurobiology and our understanding of it. In addition to developing an interface that is safe to implant and lends itself to chronic stimulation, scientists are tasked with figuring out and mimicking the stimulation patterns of a visual system that we have not fully mapped out.

#Light at the end of the tunnel?

Fortunately, there are a variety of new techniques on the horizon. Non-electrode stimulation methods such as photoactive nanoparticles allow for safer surgery while allowing more of the retina to be stimulated. Developers of vision implants are paying more attention to transmitting dynamic information via spike timing, and technology that claims to have cracked the retinal code is finally in human clinical trials. With these developments and the recent boost in BCI research and funding, there's reason to be optimistic that bionic vision will become a reality this decade.

Why Model Systems are Indispensable for BCI Research

Tue, 18 Aug 2020 00:00:00 GMT

From neurons in dishes to behaving insects to primates, how do we trade off accessibility and iteration speed against clinical relevance? What model is best suited to which research question?

Photo a micro-electrode array and a neural culture from Potter et al.

For newcomers to BCI research, it might feel like no man's land between Neuralink's implantables and Emotiv's (or openBCI's) wearables. But there are many ways to contribute to BCI, or to learn about the field.

There's a lot that happens at the interface between the wet grey gloop of neurons and the ever-finer wires that prod them listening for signals to pass outwards. The number of components that have to be developed, tested, and perfected before you can build a functional BCI is immense.

This is where model systems come in. The better we understand and develop these components — basic neurobiology, materials and biocompatibility, imaging techniques, stimulation and recording methods, and signal decoding — the faster and safer the path to BCI.

Scientists have done fascinating work with two-dimensional cultures of neurons. In one of the more well-known applications, a group trained a cultured dish of rat neurons to fly a plane in simulations of various weather conditions. Gimmicks aside, growing rat or human neurons on multielectrode arrays provides a way to study the activity of neural networks — albeit artificial ones. A neuronal culture is an elaborate and spontaneously active living neural network, but it is not likely that its organization has any correlations with real-life networks. The neuron culture technique has been standardized down to kits, with open-source software to help analyze the recordings.

This paradigm even has direct commercial potential, with 'wetware' startups like Koniku using these little dishes of lab-grown neural networks for odor detection for security, military, and agricultural applications.

Neurons grown in 3 dimensions using 3D scaffolds or a gel matrix allow even more surface area for recording. This is where these cultures start to be called 'brain organoids', bringing to mind science fiction scenarios. Every model system raises a new set of ethical questions, and these 'brain organoids' come with a particularly large list of issues to consider before you embark.

Human midbrain organoids from Sun et al. 2018

Some techniques that are relevant to today's BCI have been developed in the so-called 'lower animals', notable voltage-sensitive dyes that allow imaging of neuronal activity. Nematodes (Caenorhabditis Elegans) are 1-mm-long worms whose nervous system is completely mapped out -- providing a system to study a rudimentary yet complete network of neurons. The Openworm project is an effort towards creating a fully virtual nervous system. Leeches have also been used for similar experiments, for their accessible, simple, and well-mapped out nervous systems.

Insects models of neuronal stimulation are perhaps more captivating. Backyard brains offer a Roboroach set for high-school-level science experiments. In research labs, simultaneous multiple motor neuron activity has been studied in tethered cockroaches, but it is unclear how much of this research can be translated to humans. Other insects like bees are also been used to study the relationship between behavior and neuronal activity, primarily odor.

The RoboRoach Bundle from Backyard Brains

Rodent models are usually the first stop for testing invasive BCI prototypes— whether they are head-mounted recorders, or sono-optogenetics stimulation systems. Systems that are too large for rodents, like vascular stent electrodes are typically tested in sheep.

Research in primates is pre-clinical. Primate studies are one step away from human implantation, so implants have to be in near-final shape before they reach this stage. Human BCI aside, many seminal demonstrations of motor control, have been done in primates and much of our understanding of the motor and sensory cortices come from research in monkeys.

###Model systems relevant to BCI research

Model System	Applications	Relevance to BCI research	Limitations
Single-unit neuron recordings	Study of cell- and molecular-level neurobiological processes in concert with electrical activity.	Fundamental method of recording neuronal activity and form the basis for classic experiments underpinning the neuron doctrine.	Single-neuron recordings are a ‘method’ rather than a ‘model system’
2D neural cell cultures	Study of neurons acting as a collective.	Computational neuroscience starts to get interesting at this level Testing for material biocompatibility, co-adaptation of new materials with neural tissue Early R&D on new stimulation/recording techniques like optogenetics	The organization of an in vitro networks is very different from real-life neural networks. The effects of any perturbations, whether genetic, pharmaceutical or even electrical stimuli are not translatable
3D cell culture	More intricate and longer lasting than 2D cultures, with more neural cell types in the mix	The 3D structure allows more complex versions of the studies done with 2D cultures as there is more than 1 plane/surface for recording	Start to capture popular imagination as ‘brain organoids’, even though there are few similarities to the real brain.
Slices of brain tissue from animals or even humans (ex vivo)	Study of intact neural circuits. Each neuron in the circuit can be studied.	Most useful for studies of the hippocampus, retina or other areas with a known architecture, where many relevant cells circuit can be found in the slice	Only preserves connections along one plane. Important input connections may be lost, forming an incomplete picture
Invertebrates (Nematode worms, leeches, etc.)	Contribute to single-neuron studies or simple networks of a few well-characterized neurons operating in loops	Early R&D on new stimulation/recording techniques like optogenetics Have been instrumental in the development of techniques such as voltage-sensitive dyes	Nervous system is highly accessible, but neuronal behavior is significantly different from humans, research is rarely translatable
Insects (Arthropods)	Most complex models that are available to DIY-ers, or science labs without institutional oversight. Allow studies in ‘behaving’ organisms	Have been used to study the relationship between behavior and neuronal activity – (primarily odor) and stimulus-driven changes. Basic experiments on motor behavior, as insects walk while tethered	Nervous system is highly accessible, but neuronal behavior is significantly different from humans, research is rarely translatable
Rodents	Probably the most widely used model system for testing BCI	Proof of concept/feasibility studies for BCI prototypes, research into materials and stimulation/recording techniques Pre-clinical studies of safety and recording bandwidth Practically every BCI implant is first tested in rats – Utah array, Neuralink and Paradromics (links) Also used for spine and motor cortex interface research	Limited utility for the study of BCI function. As rodent models are relatively accessible within academic settings, neuroscience has a history of over-reliance on these for pre-clinical research, often spending graduate student time on developing hypotheses that are not translatable.
Mammals	Sheep are large enough to fit bulky prototypes	Proof-of-concept to demonstrate recording bandwidth, and long-term studies for of durability Pre-clinical safety studies	Limited utility in the study of BCI functional efficacy
Non-human primates	Precedes human implants, allows studying BCI function through 2-way communication	Pre-clinical studies of motor and sensory cortex, potentially other brain areas	Primates are not used for research until an advanced, functional interface is developed

Our understanding of neuroscience is inseparable from the models and measures we choose. If we look at single-cell recordings we're likely to think on the scale of individual neurons. If we use EEG headsets, we're going to think at the level of surface-accessible brain rhythms and terms of large brain regions. Models dictate the speed at which we iterate, and therefore the rate at which we progress.

Although none of these models will independently amount to a useful BCI, we need them to make parallel progress on all the components of an interface. The BCI community is growing, and each model system is a way for new researchers to get a foothold in the field.

How to Build a Neural Interface: Lessons from Cochlear Implants

Sat, 08 Aug 2020 00:00:00 GMT

The cochlear implant is a miracle that deserves a place in the history of brain-computer communications, as the first neural interface to restore a human sense through electrical stimulation of neurons

Photo by Michael Jasmund

How many people have an invasive neural interface implanted right now?

Hundreds of thousands. Close to a million people with hearing loss today probably have a cochlear implant to help them hear.

Cochlear implants are the most successful neural interface to date by a wide margin. Since the US FDA cleared the first cochlear implant for adult use in the mid-1980s, these devices have steadily become more and more sophisticated. At a time when no fewer than 50 ventures are building versions of brain-computer interfaces (BCIs) --- both invasive and non-invasive --- it's worth taking lessons from the neural interface that was far ahead of its time.

This post is not about the impact of cochlear implants on people's lives, whether they are always desirable for those who are eligible for them, or the difficulties of adapting to these little devices that stimulate your auditory nerves. There is probably also a debate to be had about when cochlear implants cross the line from replacement into augmentation. The day when recipients of cochlear implants may have better hearing abilities than the un-implanted may not be too far in the future.

This post is about how this first and most successful of neural interfaces was developed. Why has it been so hard to replicate this success for other senses, or brain cortical stimulation? Are there any pieces of this success story that we can replicate for invasive BCI development?

What is a cochlear implant?

A cochlear implant bypasses damage or lack of development of the middle or inner ear to directly stimulate auditory nerves and restore hearing. They are so well-known today, that we don't even think of them as invasive neural interfaces. Cochlear implants are surgically implanted electrodes that augment a sensory function by directly stimulating neurons. They have an external speech processor that hears sounds and converts them to electric signals. These signals are transmitted to the cochlea to stimulate auditory nerve fibers, bypassing the sensory portions of the ear.

The neuroscientists who developed these implants decades ago managed to mimic the tonotopic organization of receptor cells in the cochlea to build a multi-electrode implant where the spatial placement of these electrodes encodes frequency. More on this further down. Thirty-five years later, we are still far from demonstrating this level of sophistication for neuronal stimulation of other sensory or cortical areas, even experimentally.

Cochlear implants are not generally considered Brain-Computer interfaces. While the speech processor onboard the implant can be considered a 'computer', current cochlear implants do not record neuronal activity. They stimulate fibers of the auditory nerve, which is a specialized nerve located within the skull but is not part of the brain.

These interfaces are still a brilliant demonstration of how BCI technology progresses bit by bit, built on a series of advances in our understanding of human biology and how to couple it with electronics. It's also a testament to the audacity that allowed its developers to imagine that you could pass electrical impulses to a human ear and enable lost hearing.

What is special about the auditory neural interface?

A cochlear implant is a long electrode threaded into the bony spiral called 'cochlea' in the inner ear. The mammalian cochlea is named after snail for its shape. The cochlea contains a natural neural interface where 'hair cells' transform mechanical sound vibrations into electrical impulses that are passed to auditory neurons. This interface is organized tonotopically --- each location along the spiral is preferentially activated by a specific frequency range. The cochlea effectively decomposes sound waves into their component frequencies -- acting as a biological Fourier transform! This relationship between pitch or frequency and location in the cochlea is elegantly described on the cochlear implant manufacturer MED-EL's blog.

Frequency coding in the cochlea

The multielectrode cochlear implant mimics both characteristics of the biological interface. It not only transforms sound waves to electrical signals but also breaks them down spectrally in a way that approximates the natural biology. Current cochlear implants have 12 to 22 electrodes that cover the normal range of speech sounds. They do not provide a fully normal hearing experience, but the brain makes up for some of the lack of resolution and distorted input. In the months to years after receiving an implant the recipient's ability to interpret speech typically improves through brain plasticity.

Of course, the neurobiology of hearing goes far beyond a single spatially organized interface. For example, the sensitivity of the cochlea is modulated dynamically by descending nerve fibers from the brain stem. This protects the system against loud sounds and helps relevant sounds in a noisy environment. At an even higher level, the brain uses the listener's intent (attention) to determine which speaker's voice is more important and adjusts the processing accordingly.

And synthetic implants are not far behind. In 2020, cochlear implants are starting to become 'hearables'. One manufacturer advertises an implant with an audio processor that can detect when you're driving and activate the speech tracking feature to selectively listen to passengers behind you. It even has an attachment that streams music directly to the implant via Bluetooth. As their capabilities grow, cochlear implants will likely mimic more functions of the inner ear and auditory nerves and auditory cortex, while packing more computing power into their processors. They will eventually evolve from neuro-prosthetics to full BCIs -- and not at a snail's pace.

How are cochlear implants so far ahead of other neural interfaces?

Work on cochlear implants started with single electrode stimulators in the 1960s, evolving to their current multielectrode form by the late 1970s. Since then, they have continued to improve every year through the addition of transcutaneous communication, miniaturization of components, and importantly, advances in speech processing algorithms.

There are at least 5 current commercial manufacturers of cochlear implants. Three of these are approved by FDA for use in the US. Cochlear implants are covered by health insurance and by many nationalized healthcare systems for eligible individuals.

What has allowed us to make so much headway with these particular devices?

The interface between the sensory cells and the nerves is surgically accessible -- but no more so than touch receptors and sensory nerves, or the retina and the optic nerve. The cochlea is unique because the spatial representation of tones or frequencies is also represented at the same interface. This allowed scientists to understand how it works and to create a synthetic version that coarsely mimics it.

The visual system and other senses also have a similar spatial organization to encode different types of stimuli. However, for the visual system, this spatial organization happens higher up (or further inside) the brain, rather than at the same accessible surface as the sensory receptors. Since we cannot fully access or even understand this representation, the development of an efficient visual interface has been more difficult.

Much of the recent firepower in the BCI area has been directed towards recording rather stimulating neurons. Even so, there is much to learn from the story of cochlear implants.

Whether its ethical questions about why we should or should not use an implant, techniques to encourage neurons to grow around with the electrodes, or just the marvel of biology as an inspiration --- cochlear implants may have some answers.