Sign in Subscribe
By petr zhegin in astrocyte* bci newsletter —

Neurotech monthly. January 2024

Temporal interference, endovascular fUS, transparent electrodes, new implantation approach, organoids, transformers and RNNs. Does Silicon Valley rule neurotech? Onera, NeuroBell, Motif Neurotech, and others raised capital

Hettick et al., / Precision Neuroscience, 2024

Hello!

January 2024 Astrocyte* newsletter is here.

Below are also a few things I'm working on. Subscribe to get these notes and a monthly newsletter into your mailbox.

Soon to be published/what has been published:

  1. (new) A summary/reflection note on the Royal Society Neural Interfaces Summit 2023 - expected in November - published in March 2024;
  2. (new) Investing in the Brain: a Generalist Investor Guide to Spotting Winning Neurotech Startups - a guest post for NeuroTechX Content Lab published in February 2024;
  3. A note on Synchron, an endovascular brain-computer interface - November 2023 - moved to April 2024;
  4. An updated version of the BCI startup landscape - December 2023 - moved to May 2024.

Please share and subscribe if you liked it.

Subscribe

Table of Contents

I. Reading List

1) Techstack/Hardware/Wetware

  • Non-Invasive - temporal interference, electromyography, diffuse correlation spectroscopy.
  • Invasive/Minimally Invasive - endovascular fUS (whoa!), microelectrocorticography, neurobus for flexible electrodes, transparent graphene microelectrode arrays, surgical robotics, and electrode implantation using sagittal saw blades.

2) Data Science - image-guided programming of DBS, EEG transformers, RNNs for movement decoding, a review of AI in neurology.

3) Misc Neurotech Reading - electordes to record from neural organoids.

II. Startup/Corporate News

1) Business Reading - two pieces that dig into the competitiveness of  Silicon Valley neurotech startups and Neuralink in particular, neurotech and pharma.

2) Funding - capital is raised by startups that operate in various modalities, such as EEG, DBS, epidural cortical stimulation, visual and auditory stimulation, peripheral nerve stimulation, and others.

Twitter handles of the authors provided after some papers are not in a particular order; based on my ability to identify authors by handles, let me know if I missed anyone.

I. Reading List 📚

Techstack/Hardware/Wetware 🧠💿🧫

Non-invasive 🎧

Multipair phase-modulated temporal interference electrical stimulation combined with fMRI. Temporal Interference stimulation (TI) combines the depth characteristic of deep brain stimulation (DBS) with the non-invasive nature of transcranial electric stimulation (studies done on the hippocampus and striatum, for example). However, there is a risk of off-target stimulation. This study focuses on mice's prefrontal cortex (PFC) and develops a computational modelling approach to identify the best electrode locations for targeting PFC. Here is the proposed solution:

To minimize the occurrence of off-target effects, we propose a new configuration with three electrode pairs, one of which has an active phase shift of 180 degrees in relation to the other two TI-inducing pairs.

X: @Val_Beliaeva, @Valerio_Zerbi, @RafaPolania, @m_markicevic, @RazanskyLab, @ViktorJirsa

Improved temporal and spatial focality of non-invasive deep-brain stimulation using multipolar single-pulse temporal interference with applications in epilepsy. This paper addresses other challenges of temporal interference, namely: weaker electric fields at target, compared to TMS, for example; inability to stimulate via controlled bursts, so it is difficult to predict when the stimulation will have its intended effect; limited control over electric field intensity. The following were studied in mice:

we replace the AM signals of TI with phase-shift keying (PSK) modulated signals, giving superior control over the modulation envelope slope, the pulse duration (independent of the stimulation frequency), and the ability to provide classic bursts (single pulse – spTI). ... We then add a multipolar configuration (multiple pairs of stimulation electrodes - mpTI) to create significant increases in the electric field at the targeted brain structure while avoiding a corresponding increase elsewhere (or even reducing off-target exposure)... Finally, we apply Fourier components in the multipolar layout to replicate classic square biphasic bursts of square pulses – transforming the aggregate pulse into a square pulse, fully replicating classic DBS, but using transcranial electrodes.

@AcerboEmma, @ViktorJirsa, @GrossRobertE, @eglo_physchem

Decoding hand and wrist movement intention from chronic stroke survivors with hemiparesis using a user-friendly, wearable EMG-based neural interface. Here is a proposed electromyography-powered interface:

... an investigational device which consists of a wearable forearm sleeve with 150 embedded electrodes and associated hardware and software to record and decode surface EMG. Here, we demonstrate accurate decoding of 12 functional hand, wrist, and forearm movements in chronic stroke survivors, including multiple types of grasps from participants with varying levels of impairment.

X: @BryanSchlink, @ianbaumgart

Non-invasive low-cost deep tissue blood flow measurement with integrated diffuse speckle contrast spectroscopy. The Diffuse Correlation Spectroscopy (DCS) method for deep tissue blood flow measurements is not widely adopted due to hardware requirements. DCS estimates blood flow from temporal fluctuations of diffuse photon intensities detected from the tissue surface. To ease these requirements, it was demonstrated that '... that fast blood flow measurements can be performed without loss of signal-to-noise using generic single mode diode lasers placed directly onto the probe, i.e., illuminating the tissue without a fibre optic cable'. This paper takes it further:

Here, we have significantly relaxed the hardware requirements for the detector and show that deep-tissue blood flow measurement is possible with a low-cost photodiode... Instead of using a SPAD [Single Photon Counting Avalanche Photodiode ] array or a camera, iDSCS detects light with a single photodiode with a custom integration circuit. 

Invasive/Minimally Invasive 🧑‍⚕️ 🏥

Invasive vs. non-invasive neuronal signals for brain-machine interfaces: Will one prevail? A 2016 overview of EEG-powered BCIs vs. invasive modalities. Includes a detailed explanation of how modalities differ based on their ability to capture the number and type of neurons, signal composition, and spatial distortion. This opinion concludes:

...once technical, socio-ethical, and neuroscientific challenges are resolved, user concerns might subside, and invasive BMIs (using primarily intracortical and potentially epicortical recordings) will prevail in most applications; certainly those for restoration of motor functions and perhaps even in applications not medically indicated.

Note that this paper does not cover other non-invasive/less invasive modalities, e.g. MEG or fUS. It was also written before the proliferation of foundation models, which are now applied to EEG signals.

Ingenious brain stimulation research offers treatment hope for neurological diseases (fUS + endovascular) - an exciting example of two technologies with high platform potential combined: focused ultrasound and endovascular delivery. Synchron, an endovascular BCI developer, collaborates with a team from the University of Melbourne. This is how the system will look like:

For the ultrasound stimulation, an array of tiny transducers would be inserted permanently into a blood vessel within the brain, most likely via the jugular vein, similar to how stents are inserted into the heart and brain today.

X: @davidgrayden, @Bionics_Sejohn

Nanoporous graphene-based thin-film microelectrodes for in vivo high-resolution neural recording and stimulation. I liked the accent on standard semiconductor industry fabrication technology and, therefore, scalability. Transverse intrafascicular multichannel electrode (TIME) and microelectrocorticography (µECoG) were implanted in rats. The electrodes themselves look exciting:

This work presents a nanoporous graphene-based thin-film technology and its engineering to form flexible neural interfaces. The developed technology allows the fabrication of small microelectrodes (25 µm diameter) while achieving low impedance (∼25 kΩ) and high charge injection (3–5 mC cm−2).

X: @damiavnc, @EduardMustbe, @xaviernavarro8, @VilaplanaUB, @naolmu, @SMarti_Sa, @chiaraspadar, @vickypuig

Scalability takes a central stage in another paper - Flexible, scalable, high channel count stereo-electrode for recording in the human brain. The team developed the Micro-stereo-electro-encephalography (µSEEG) system:

... monolithically integrated human-grade flexible depth electrode capable of recording from up to 128 channels and able to record at a depth of 10 cm in brain tissue. This thin, stylet-guided depth electrode is capable of recording local field potentials and single unit neuronal activity (action potentials)...

Scalability is achieved via:

... (1) titanium (Ti) sacrificial layers employed in the microfabrication of free-standing microelectromechanical system (MEMS) devices. ... (2) This MEMS process is implemented on relatively large (18 × 18 cm2) glass substrates allowing us to produce multiple copies of the SEEG devices using materials that are typical for the manufacturing of display screens. ... and does not involve manual assembly typical for standard SEEG electrodes.

Here is a summary of the paper and commentaries from the first authors.

X: @ACPaulk, @Biologish, @jpezaris, @JihwanJamesLee, @RMarkRichardson, @raslanneuro, @SBenHaimMD

Thin-film can be used not only for electrodes, but to host multiple components of an implant. NeuroBus - architecture for an ultra-flexible neural interface presents a '... distributed direct digitizing neural recorder ASICs on an ultra-flexible polyimide substrate are connected in a bus-like structure...'. This architecture allows:

...short connections between electrode and recording front-end with low wiring effort and high customizability. The small size (344 μm x 294 μm) of the ASICs and the ultraflexible substrate allow a low bending stiffness, enabling the implant to adapt to the curvature of the brain and achieving high structural biocompatibility.

A rodent animal model was used to validate 'the joint capability of the recording front-end and thin-film electrode array'.

X: @ilka_diester, @AdzemovicAhmed, @ilka_diester

High-density transparent graphene arrays for predicting cellular calcium activity at depth from surface potential recordings. You probably noticed how frequently I refer to platform station, standardisation, and scalability. These features were fundamental to the success of the biggest tech companies; e.g., without Wintel, the success of PCs would hardly be possible. The tech ecosystem tends to converge on the dominant technology. However, human brains are more complex than computers. Here is how one might think about a bottleneck to the dominance of a single modality/tech in neuro:

Spatial scales encompass neural circuits in millimetres or centimetres, single neurons in micrometres, synapses in submicrometres and proteins such as ion channels and receptors at the nanoscale. This spatial diversity also cultivates temporal diversity where some molecular processes are taking place in microseconds, action potentials in sub-milliseconds, neurotransmitter or hormone release in minutes to hours and learning and behavioural changes in hours to days. Monitoring neural dynamics and interrogating neural functions across these diverse spatial and temporal scales is not possible using a single tool or technology. 

A proposed solution is:

[we] ... demonstrate completely transparent, high-density, high-channel-count (up to 256 channels) microelectrode arrays with ultrasmall graphene electrodes for multimodal experiments. 

These arrays were implanted over the visual cortex of mice.

Transparent electrodes allow to see deeper into the brain while remaining less invasive (from a great write-up by Michael Franco):

This allowed the researchers to simultaneously shoot lasers through it [electrodes] and use a two-photon microscope to image calcium spikes from neurons that were up to 0.25 mm below the surface. Calcium is a key component of the way in which neurons transmit data to each other.
The researchers were then able to train a machine learning model to establish the link between surface activity and sub-surface activity, in effect teaching it to understand what is happening deeper in the brain based on what the sensor picks up from surface signals.

X: @DuyguKuzum, @takaki_komiyama

Implants need surgery. And surgeries must be automated to some extent to make neurotech scalable. I feel that the robotics component of neurotech is frequently overlooked, e.g. just a few startups are working on robotic surgeries. Ironically, brain biopsy was an application for the first surgical robot, so robotics is in neurotech's DNA. The IDEAL framework for surgical robotics: development, comparative evaluation and long-term monitoring '... presents practical recommendations to guide robotics developers, clinicians, patients and wider systems from multiple perspectives - including economics, surgical training, human factors, ethics, patient perspectives and sustainability.'

X: @hani_marcus, @dzkhan94, @hugo_lh, @JohnHanrahan1, @david_j_beard, @KenCatchpole, @MaroeskaRovers, @NicholasRaison, @prokarurol, @StockenDeborah, @McCullochP

In addition to automating surgeries, making them less invasive is required. Minimising invasive is among the core themes in The Layer 7 cortical interface: A scalable and minimally invasive brain–computer interface platform. The paper describes microelectrocorticography (µECoG) system that represents a combination of minimal invasiveness and improved signal quality:

Two versions of the microelectrode array were fabricated for this study. The first comprises 529 electrodes of multiple sizes ranging in diameter from 20 to 200 µm, while the second version comprises 1024 channels of three different diameters (977 at 50 µm and 42 at 380 µm, with an additional 5 reference electrodes at 500 µm). Microelectrode arrays can be inserted individually or in modular assemblies...

The system is wired, where...

... each array [is] connected to a customized hardware interface. After subdural array implantation, the interconnecting cable of each microelectrode array module passes through a dural incision and a cranial micro-slit incision, is tunneled under the scalp as needed, and is connected to an individual head stage.

The surgery looked as follows:

The procedure employs precision sagittal saw blades to make 500-to-900 micron-wide incisions in the skull at approach angles approximately tangential to the cortical surface, facilitating subdural insertion of our thin-film arrays without requiring a burr hole or craniotomy...

In vivo testing of the minimally invasive surgical insertion technique and electrode array performance were performed in adult female Göttingen minipigs. Human intraoperative array implantation was performed after traditional craniotomies to expose the regions of surgical interest.

X: @kate_gelman, @VanessaMTolosa, @PeterKonrad12, @cmermel

Data Science 💻🧮

The use of image guided programming to improve deep brain stimulation workflows with directional leads in Parkinson’s disease. My intuition is hardware is still the bottleneck in neurotech. Here are the results from a trial that aims to explore to what extent software, specifically image-guided programming (IGP) software, improves BDS and, therefore, in my opinion, leads to wider adoption. Results:

We observed 78% concordance between the two electrode levels selected by the blinded IGP prediction and CSC [clinical standard care] assessments. In 64% of cases requiring refinement, IGP improved clinical efficacy or reduced mild side effects, predominantly by facilitating the use of directional stimulation (93% of refinements).

EEGFormer: Towards transferable and interpretable large-scale EEG foundation model.

In this paper, we introduce a novel EEG foundation model, named EEGFORMER, for self-supervised learning using large-scale EEG data. Our approach leverages a vectorquantized learning algorithm to simultaneously learn a discrete codebook and representations of multi-variate EEG signals.

X: @SongKaitao

Deep learning in EEG-based BCIs: A comprehensive review of transformer models, advantages, challenges, and applications. An amazing overfiew I'de suggest every startup founder conteplating an EEG-powered startup should read. It reviews dozens of papers on transformer architectures aplied to motor imagery decoding, emotion recognition, and sleep stage. It also points out emerging applications, e.g. seizure detection, speech reconstruction, etc.

Brain control of bimanual movement enabled by recurrent neural networks focuses on applying recurrent neural networks (RNNs) for decoding multi-effector motion. It concludes:

... neural network decoders may be particularly well-suited to the problem of decoding multi-effector motion due to the nonlinear structure of the neural code associated with such movements. We show that it is possible to enable simultaneous control of two cursors using recurrent neural networks with good performance, if care is taken to train them in a way that enables successful transfer to online control.

X: @WillettNeuro, @shenoystanford, @neuroleigh

Artificial intelligence in neurology: opportunities, challenges, and policy implications reviews 66 original articles and explores the value of AI in neurology and brain health. I particularly liked the relational graph between main data modalities (medical imaging, non-imaging biomarkers, BCIs, etc.), types of AI (foundation models, simulations, etc.), and clinical applications (prevention, early detection, etc.).

X: @mario1geiger, @SebastianWinter, @pkarschnia, @EJVaios, @IraHaraldsen, @ValeryFeigin, @pswieboda, @vivnat, @NitaFarahany

Misc Neurotech Reading 👨‍💻

Kirigami electronics for long-term electrophysiological recording of human neural organoids and assembloids. Neurotech faces not only engineering but also research challenges. Too little is known about neural development and early drivers of conditions like autism or epilepsy. A novel technique for growing 3-D clusters of human brain cells (brain organoids) in Petri dishes emerged to propel research. Then, a challenge followed of how to collect data from organoids (from an article by @grace_huckins):

Traditional recording techniques are designed to measure brain activity in living animals or in cells cultured on a fixed surface. Brain organoids, however, float freely in fluid and are easily damaged by recording devices. And since organoids are much smaller than an entire brain, even a little damage can be catastrophic.

To address this, a research group came up with the following system, described in a paper:

 ... flexible electronics that transition from a two-dimensional to a three-dimensional basket-like configuration with either spiral or honeycomb patterns to accommodate the long-term culture of organoids in suspension. ...this platform, named kirigami electronics (KiriE), integrates with and enables chronic recording of cortical organoids for up to 120 days while preserving their morphology, cytoarchitecture and cell composition.

X: @XiaoYang63, @CsabaForro1, @Sergiu_P_Pasca, @santorof14, @BianxiaoC

Startup/Corporate News 📰💰

Business Reading

Perspective by Dr. Kip Ludwig: What does the recent Neuralink FDA IDE really mean? - A dive into where Neuralink's competitive advantages might be compared to those of other leading BCI developers. Among these advantages are:

1) More electrodes for recording; 2)Wireless; 3)Very small thin-film electrode arrays to minimize disruption of tissue, hopefully leading to better recorded signal across years of implantation; 4) A novel ‘robotic sewing machine’ to get lots of very small electrode arrays into the brain very quickly; 5) An entirely in house supply chain. This in theory would give Neuralink more control over the full final product...

I bet that #4 and #5 are the most critical ones. #4 affects scaling/growth, while #5 allows them to iterate faster on product development and create a 'premium' product.

Another interesting idea from the article is the need to complement innovators/disruptors with professionals from incumbents.

... Musk is deliberately avoiding people with that experience [who can navigate a complicated regulatory space] by design to avoid ‘contaminating the thought process'. This is unbelievably stupid and very concerning

I think hiring from incumbents makes sense, but one should be careful with this as cultures clash. Amazon, whose founder Jeff Bezos admired Walmart, a competitor, started to hire from there around 1998 when Amazon's culture was settled (it was founded in 1994). In 1998, Amazon had 2.1K employees, while Neuralink had 443 employees, according to LinkedIn in March 2024. It might be too early for Neuralink to hire senior leaders from outside.

What Silicon Valley needs to know about the brain, and what it doesn't — a fantastic essay on how Silicon Valley (SV) startups may move the needle in neurotech. @MaromBikson highlights several leavers that startups could pull (hardware, e.g. by increasing channels, robotics; and software — by leveraging machine learning) and asks:

How then will Silicon Valley, in a short time-span and by orders of magnitude, exceed the performance of existing devices developed over decades with billions of dollars from governments?

Marom mentions several avenues how startups could succeed, including brute force over-engineering of brain technology (using increasingly dense hardware), cracking open these mysteries of the mind (and building very targeted tech based on it), and coming up with a relatively reduced product, kind of a '... killer app that still avoids the need for precise reading or writing the brain – perhaps something that acts more like power-steering than a self-driving car.'

I agree with Marom that all ' ... these scenarios are at least plausible.' However, I think the winning company or companies will probably be built based on Silicon Valley entrepreneurs' ability to platforms existing technologies, i.e. enabling feedback loops and network effects that make a platform more significant/more valuable than a sum of its parts.

I wrote extensively on software platformisation (here for data startups and here for foundation model developers), and I'm sure platformisation will be the key to the success of a neurotech startup. Here is a description of how the platform approach works for Space X, another hardware-heavy, deep tech, though not a neurotech startup:

This platform approach to rocket-making creates a virtuous circle. Rocket systems made up of modular components are more easily upgradeable and reusable. This results in an increase in volume – in this case of launches. As people upgrade and recombine the components of their platform (the rocket) they can repurpose it, while continuing to scale. The variety creates the conditions for more scale, because it means that the platform has more value to more users.

This research summarised Space X's platform approach via several characteristics: '... Repeatability, extendability, the ability to absorb new knowledge and adapt to new situations.'

Connecting multiple technologies in one platform vertically (an implantable device and a surgery robot), horisontally (multiple modalities working together or in a modular fashion to address multiple indications) or employing marketplace components (like engaging with pharmaceutical companies/enterprises for monetisation and subsidising consumer products) are among noticeable approaches for neurotech.

The ability to build platforms that tap various technologies, markets, indirect monetisation routes, etc., is a definitive feature of startups that academic institutions and corporate incumbents are less prone to. The platform approach does not necessarily require startups to produce breakthrough research; it focuses on ways in which various discoveries/technologies could be combined and deployed in a scalable way.

Another feature of Silicon Valley entrepreneurs is their ability to raise funding for startups that address customers' unnoticed and somewhat 'excessive' needs. Think of Twitter; who would think about a 'need' to have such a counterintuitive social platform?

While researchers and medtech incumbents focus on precise medical needs (here is a list of 35 ongoing clinical trials exploring future DBS indications, for instance), SV founders could experiment with applying neurotech to lucid dreaming, spiritual experiences, and other applications that governments and traditional biotech investors could hardly fund.

X: @MaromBikson

No moonshots with paper planes: Alzheimer’s trials need neurotech - this article highlights neurotech potential for trials, specifically for Alzheimer’s

Life sciences executives are clamoring for more efficient trials, new real world evidence, and better, objective (and I’ll add, patient reported) measures of cognitive decline, drug efficacy, and health. I see this as writing on the wall that 2024 will be a bumper year for neurotech startups selling into the life sciences companies.

While I agree with the potential and the need, I suggest neurotech founders be careful when turning their heads to pharma, particularly if it was not an original intention. From an investor perspective, the biggest outcomes are frequently associated with businesses that capture the entire value chain rather than servicing specific players across it (they could be first customers, though).

In deep tech, examples of serving the entire value chain are Waymo, a self-driving car developer that operates its own ride-hailing service, and Space X, which provides space launch for customers and is tightly integrated with Starlink, a satellite service. 'SpaceX’s rapidly growing fleet of Starlink internet satellites now make up half of all active satellites in Earth orbit.' - a research says.

So, if your aspirations are to create an end-to-end offering, try not to be distracted by opportunism.

X: @naveen101

Funding

🇨🇳 BrainThink raised an undisclosed angel round. The company integrates wearable sensors to deliver scalp electroencephalogram (EEG) and heart rate variability (HRV).

🇮🇪 €2.1m investment in Irish medtech spin-out NeuroBell. NeuroBell has developed an easy-to-use, pocket-sized wireless brain EEG monitor which can detect seizures in babies.

🇳🇱 Onera raises over €30M in series C funding. Onera offers remote sleep diagnostic and monitoring solutions, enabling clinicians to conduct sleep studies. Among other things, the system records electroencephalogram (EEG), electromyography (EMG) to measure upper airway muscle activity and electrooculography (EOG) to register eye movements.

🇮🇹 Newronika secures €2 million in blended finance from EIC accelerator. Newronika’s adaptive deep brain stimulation (DBS) system for Parkinson’s disease is designed to adjust therapy throughout the day based on individual patient needs.

🇺🇸 Motif Neurotech closes series A financing of $18.75 million.  The company develops wireless, minimally-invasive neuromodulation therapeutics for mental health. The company's lead product is a miniature implant in development for treatment-resistant depression. Based on a preprint, it probably employs electric epidural cortical stimulation (ECS).

🇺🇸 Cognito Therapeutics raises $35M series B extension. Cognito builds a device that delivers both visual and auditory stimulation for the treatment for Alzheimer’s.

🇺🇸 Nalu Medical, Inc. announces $65 million equity financing. It offers peripheral nerve stimulation (PNS) against chronic pain.

🇺🇸 Rune Labs unveils $12 million strategic round. Rune Labs’ software ecosystem for Parkinson's disease collects patient symptom data, including tremors and dyskinesia, through measurements made by Apple Watch.

📬📬📬 Subscribe for a monthly update on neurotech and neurotech investment/commercialisation activity.

Subscribe

Subscribe to peter zhegin | future platform

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.
jamie@example.com
Subscribe