Electrodeless Neuro-Interfaces
  • TheGreyKnightTheGreyKnight November 2015
    As of now, there are a lot of projects that use electrodes and indirect interfaces via magnets, or vibrators to supply data to the user. But there are a host of potential issues with these methods that I'd like to avoid. It's easy enough to read the current state of the brain via EEG. We can even use tDCS, TMS to change how our brains work without ever sticking an electrode below the skin. The question is, why can't we use a similar method to create a direct interface with a nerve, even if it isn't the brain?

    Any thoughts on how this could be done with precision (not just slapping an electrode on the skin are running current between point A and B to stimulate a nerve)?
  • TheGreyKnightTheGreyKnight November 2015
    I found a rather dusty paper from '65 (LINK) that said some interesting things. I'd be interested in repeating their experiments with modern techniques and kit to minimize error (and verify their claims).
  • MTSMTS November 2015
    You might've seen this video already but its very interesting what she's getting at: https://www.solveforx.com/moonshot/5631639635886080#solution

  • bciuserbciuser November 2015
    The answer to this question is one of practicality. In order to generate TMS impulses that are capable of affecting a nerve's membrane potential you need a fairly massive transient current. It's just simply not possible to generate with portable or wearable devices, much less implantable ones. The reason for this has to do with the fact that magnetic fields (which are the only way to feasibly nonlocally influence membrane potential) are weaker with distance. Not only are they weaker, they are also larger and become more difficult to control.

    If you want to affect things from a greater distance, you give up specificity and increase energy required, both of which are extremely important in any neural interface.
  • bciuserbciuser November 2015
    As for the paper, the very first sentence states why this isn't feasible for a neural interface. "Using wavelengths of the centimeter wavelength". Any waves that have a wavelength of 1 centimeter cannot be controlled on a scale that is meaningful to the nervous system (mm to nm scale), and isn't a candidate for a remotely useful neural interface.
  • TheGreyKnightTheGreyKnight November 2015
    Could you use weaker fields and pump the targeted nerve with some sort of EM Radiation to increase the transient current?(Just asking for the sake of asking, because what I know about physics dictates it probably wouldn't.)
  • bciuserbciuser November 2015
    No, you'd just make whatever particles you injected move. Any meaningful increase in conductance would require concentrations that are cytotoxic.
  • garethnelsonukgarethnelsonuk November 2015
    Why bother when electrodes work fine?
  • TheGreyKnightTheGreyKnight November 2015
    Because cutting a hole in the skull is kinda invasive. For the same reason that haptics exist. 
  • garethnelsonukgarethnelsonuk November 2015
    For anything practical, electrodes work fine.

    For high-resolution interfaces, you're not going to get anything both portable and non-invasive.
  • JohnDoeJohnDoe November 2015
    I don't think I am completely following, are you trying to connect with something telepathically, or are you trying to just avoid using electrodes? If the latter could nerve cells be engineered to make a wired connection from the brain to a port for example. Almost like a networking cable, ware only these impulses could travel along these special nerves. I am going to start reading that paper.

    John Doe

    "Because cutting a hole in the skull is kinda invasive." May I ask why you are using invasive electrodes? The only place I know for certain they are used is epilepsy research, that just caught my eye.
  • TheGreyKnightTheGreyKnight November 2015
    For lack of better examples, the supposed "microwave hearing" effect, or Nerve Gear from Sword Art Online are the best examples I can really think of for what I'm thinking of. You can think of it like telepathy, if that works for you.

      @garethnelsonuk What about non-portable systems? If a rig the size of an MRI machine is needed for the first iteration, we can always work on making it smaller later.


    Does anyone know whether the voltage you introduce has a bigger impact on the strength of the signal you introduce into a nerve, or whether it's current?
  • bciuserbciuser November 2015
    Magnetic fields are responsible for the nerve stimulation, and magnetic fields are induced by a moving charge (current).
  • TheGreyKnightTheGreyKnight November 2015
    Still has a ways to go, but something Vanderbilt was working on looked really cool: Link

    Apparently they were able to cause nerves to fire using infrared, instead of electrical stimulation. Funny enough, Infrared also happens to include wavelengths as small as, and smaller than the smallest neurons found in humans. Near infrared also penetrates soft tissue, bone, and brain matter according to this. Another interesting article on the subject: Link.

    That said, I happen to have 2 IR laser diodes(no idea what frequency or power rating). I'm going to do some investigating and do my best to get an experiment with this set up.
  • bciuserbciuser November 2015
    Unfortunately the authors seem to be silent on device design, so I'm not sure what they mean by "low cost" and "compact". Also, they stimulate the exposed nervous tissue directly. it isn't realistic to think you would be able to achieve anywhere near this resolution after dealing with the absorption and scattering of IR by the tissue in between.

    Another issue is that this is only a unidirectional technology: communicating from a device to the brain. The hemodynamic response is too slow for any real neural interfacing (10s+ delay and not consistent). Cool paper and certainly not without application, but still invasive.
  • TheGreyKnightTheGreyKnight November 2015
    Which kind of resolution are you talking about? Spatial resolution, meaning the smallest quantity of neurons the device is capable of interfacing with at a single time, Or something else?

    If spatial resolution is what you're talking about, couldn't you simply design a device that has an extremely high starting resolution, so that when your IR signals get scattered and absorbed by the tissue, the resolution is decreased to the minimum level needed for an effective neuro-interface? Or does the skin introduce a constant cap for the resolution you'll get?

    Smallest neuron size: ~4 microns(micrometers)
    Device Resolution: 2 microns(micrometers)
    Resolution Loss: 2 microns per tissue-size unit.
    Tissue thickness: 1 tissue-size unit
    End Resolution: 4 microns

    ---- EDIT ----
    I didn't think to use square micrometers when I first came up with the example, but now that I've thought about it, it would just make things more complicated if I had.
  • TheGreyKnightTheGreyKnight November 2015
    One option for the other half of the problem is EROS(Event-related optical signal). It uses infrared light to measure the optical characteristics of the neuron's themselves, rather than the flow of blood. It's non-invasive too. Here's a link to an article about it, which is unfortunately behind a paywall: LINK
    Could someone grab it, and add it to the community article library( @cassox or @glims)?
    This is another related article, also behind a paywall: LINK
    And here's one that doesn't cost anything: LINK
    Sadly, it sounds like it falls short of the spatial resolution needed for a BCI. Is "within millimeters" accurate enough, because I don't think it is?

    I also found an interesting article on Dynamic Brain Imaging, which seems to give a good rundown on the subject:  LINK
  • ElectricFeelElectricFeel November 2015

    Here, grabbed the two behind the paywall for you.
  • SlachSlach November 2015
    @ElectricFeel Bless you.

    @TheGreyKnight A spatial resolution of a few millimeters is definitely enough for a working BCI. You can even get a BCI working with EEG, so the mm-range is definitely 'deep' enough. ECoG falls within the same range. Of course, it depends on what you want to do. You're probably not going to control a prosthetic arm with 7 degrees of freedom with minimal training, but doing something like moving a cursor is definitely doable with that type of resolution. I wonder why this EROS method isn't more widely used in research. Can't be that expensive, right? Is it susceptible to signal noise or does the signal easily get messed up with small head movements?
  • TheGreyKnightTheGreyKnight November 2015
    @Slach The Problem with EROS is the fact that it doesn't go nearly as deep as MRI. Only a few centimeters. It's actually super cheap and easy to use, I believe.

    And @ElectricFeel Thank you so much!
  • bciuserbciuser November 2015
    "Better than a centimeter" does not mean "a few millimeters". The authors also say the relatively low SNR "in reality requires that data is averaged across subjects". I.e. cannot be used reliably in real-time (required by a BCI), but can be useful when studying brain activity across multiple people. While it has better spatial resolution than EEG, it has much worse temporal resolution (millisecond for EROS vs microsecond for EEG). 

    In terms of penetration, most of the interesting cognitive stuff is happening in the neocortex (the outermost layers of brain tissue), so the 3-5cm penetration limitation isn't a huge issue.
  • TheGreyKnightTheGreyKnight November 2015
    As long as the millisecond delay is under 50 ms, it's not a big deal, because there's already a 50 ms delay between visual stimuli and auditory stimuli. By that benchmark, it's definitely an acceptable delay.

    In regard to the SNR, do you or @thomasegi know about how to improve that ratio?
  • bciuserbciuser November 2015
    No, that's not exactly how it works. The way EEG activity is typically made meaningful (and really all non-spike neural data) is by sorting the data into frequency bins and plotting the intensity at those bins over time. This is called a spectrogram. If you are only recording at 20Hz, you can only capture 10Hz signals in a perfect world, more realistically less than that. To capture useful data you need to record at higher frequencies (100Hz+). If you are recording individual neurons the story is very different, but that's not an option with this technology.

    No, I don't know how to improve the SNR here.
  • TheGreyKnightTheGreyKnight November 2015
    And here's the study on EROS you've all been waiting for. No more "better than a centimeter", or other imprecise nonsense. LINK Here's another one with meaningful data on about and EROS experiment and the configuration used: LINK
    This article(LINK) mentions a more recent study published in the Journal of Cognitive Neuroscience about using EEG and EROS signals together. I'd like to get a copy and see if they mention exactly how they configure their experimental apparatus, and how they used the 2 data sources in conjunction.
  • JohnDoeJohnDoe December 2015
    This may also be worth taking a look at. I was looking at options other than EEG for my project, just in case for what ever reason it proves to be impractical or what ever. No way in hell with current technology could I use that for my project, but it may help you. I would also like to point out that you did say non-portable systems were a option.

    John Doe
  • TheGreyKnightTheGreyKnight December 2015
    I had a look at this when I was first diving into the subject, but the whole "shielded room" thing was a bit of a problem. Seems like it could work, but pretty expensive, and it suffers from range issues. 
    One paper mentioned something that might make it cheaper, though( Link)

    I might see about making a channel on slack about this soon.

    On another note, if someone has access to this paper (LINK), It'd be a big help if they could download it for me. 
  • TheGreyKnightTheGreyKnight December 2015
    Slack channel started. Message me if you want to be added (already added everyone from this thread, if they were on slack).
  • JohnDoeJohnDoe December 2015
    I wonder if a magnetic field could not be created around the user to drown out interference and then digitally cancelled out, my foresight tells me the immediate problem would be you would need extremely sensitive sensor’s that can detect across a extremely variable power range of fields. Correct me if I am wrong but are those fields coming off of the brain in the picoteslas? Let me make a comparison to a humbucking guitar pick up. Just in case that was unclear or vague.

    John Doe
  • TheGreyKnightTheGreyKnight February 2016
    I've made some progress in this project, but I haven't quite gotten all of my information and data into nice, readable format yet. Might be a month or so before that happens.

    Question for @bciuser , or anyone else who might know they answer:  If I've got an interface device designed to pass data into the visual cortex, how many data I/O (In the form of electrodes or laser targets) sites per square millimeter will I need? Do I need to interact with individual neurons, or will dealing with clumps of 10 - 20 be sufficient to allow a good connection? 

    Changed units from mm to mm^2.

  • JohnDoeJohnDoe February 2016
    Are you building a MEG?
  • TheGreyKnightTheGreyKnight February 2016
    Not as of now. I'm still sticking to the infrared route, for a number of reasons. Size is a big one. MEG might go somewhere in the future, but right now, the juggernaut of advantages backing optical systems can't really be beat. And they have the added bonus of being multi-purpose. I can stimulate and observe with the same laser/light source. 

    On a similar note, I have something pretty cool I'm about to test that'll be useful for this project, and eliminate the need for chemical-based anesthesia. Reusable, and all that jazz. I just need to order some parts and I'll be able to give it a test. I'll probably be making a thread on it soon. 
  • JohnDoeJohnDoe February 2016
    How are you optically reading brain waves?
  • TheGreyKnightTheGreyKnight February 2016
    So, when you shine infrared light on a neuron, if that neuron fires, it changes the optical properties of the cell membrane. That change causes it to reflect light differently, allowing you to extract data on the cell's current condition. 

    If you increase the amount of energy you're introducing to said neurons via infrared light, you can cause them to fire an action potential. I theorize you could effectively combine these technologies by using one frequency of infrared to read and pre-charge the region you'll be "writing" to, and another frequency to increase the total energy in the region to the point that everything fires. You may even be able to use the same frequency for all of it. But if you were pulsing the system on and off, you'd need to account for the spikes in energy.
  • JohnDoeJohnDoe February 2016
    Can inferred penetrate the human skull? Furthermore could enough of it be reflected back that you could varify a change in brain state? Maybe something like a photo multiplayer tube? Shielding of sensors should be a a bitch. Could run off the some of the same hardware of the ModulerEEG.... Hum I am vary interested now. You may have figured out the key to nural input.... While at the same time makeing a tDCS device 1000x safer....
  • TheGreyKnightTheGreyKnight February 2016
    Yes, some frequencies of infrared can penetrate the human skull. According to one of my sources (LINK), "the maximum optical penetration can be estimated to be about 1.5 cm when a detector is placed at 4 cm from the source." (Page 1, Paragraph 4, Line 10)

    Very, very important note. There are 2 ways of using this technology. The first is observing the vascular activity in a given region (Changes in the absorption properties of the tissue). This is okay for BCI's where time isn't an issue.
    The other option is to use it to observe the neurons themselves (Changes in the way the tissue scatters light). When a neuron fires an action potential, it changes the way it scatters the light nearly simultaneously. This in particular is what I'm interested in. If all the claims in the papers I've found are accurate, this ought to provide temporal resolution comparable to electrocorticography.

    I'd be interested to see if it's possible to use technology similar to holography to capture and utilize the data.

    Shield the sensors from what? It's just like a camera. Just with infrared light instead of visible light.

    What I haven't found yet are trials that use this technology to observe the brain while stimulating it with tDCS, or something similar. So, it's kind of hard to get a feel for how the rate at which neurons fire relate to an EEG signal. So, if anyone's interested in assisting with such a study, let me know. I'll start writing up the procedure for the tests.

  • JohnDoeJohnDoe February 2016
    So you are not worried about any IR noise messing with you sensor(s)? I will have more questions just not going to ask anything till I read that paper.

    John Doe
  • TheGreyKnightTheGreyKnight February 2016
    I'm still working on an actual design for a sensor array. As far as how I'm addressing noise, I intend to eliminate as much electrical noise as I can. External infrared noise should be very easy to block. Interplay between the sensors in the device could be a problem. I've yet to start tearing apart the finer details of the various set-ups used by the researchers for these experiments.

    The first order of business for getting a scatter-based sensor working is establishing all of the variables. Does a change in the oxygen level of the blood through a certain spot change the scattering behavior of that region? How much does the scattering coefficient of a neuron change when it fires?

    This study (LINKSOURCEARTICLE Shout out again for @ElectricFeel getting ahold of it) goes over and reviews a number of studies on EROS. It mentions the potential of scattering and absorptive, but, unless I've missed something, they don't say outright which of the 2 is being used. I'd really like to get my hands on some stuff to do a quick proof of concept.

    In regards to sourcing IR diodes and lasers, do you have a good source for getting diodes that emit IR at 690 nm @ThomasEgi ?

  • JohnDoeJohnDoe February 2016
    How many do you need and how many milliwatts do you need? Have you seen this: Thorlabs HL6738MG Laser Diode?
  • TheGreyKnightTheGreyKnight February 2016
    For proof-of-concept, just one. According to this (LINK), an LED that was only 500 microWatts was sufficient. I'll do a bit more looking, and see if 30 mW is too much. That definitely looks like it might work.
  • ThomasEgiThomasEgi February 2016
    TheGreyKnight there are many LED's around that wavelength, and almost no lasers at the exact wavelength. You can get 670nm Lasers.  If you don't have to use lasers for a very good reason, don't use them. They are super delicate, short lifetime and they pretty much die if you look at them with an angry expression on your face ( got this validated by a number of people working with solid state lasers).
    Also, for lasers you can't just scale down power as you'll have the laser threshold somewhere. As for LED's it scales pretty well with current.
    As for LED's there's plenty of choice, if you can list me some requirements they have to met I might be able to help you. Deep red can get you as close as 660nm. Ir are typically around 850nm and more. 
    My workhorse for optical communication systems are the SFH203 (and variants), in combination with regular red led.

    As for low-noise design, please hit me on IRC. I got a good share of experience with optical systems/circuits, analog and very low noise designs.
  • TheGreyKnightTheGreyKnight February 2016
    Well, is there a time you're usually on IRC @ThomasEgi ?
    And just to make sure I've got it right, #biohack goes in the channel? I'm pretty unfamiliar with IRC.

    As far as requirements, something meeting these criteria should do.
    • Wavelength: 690 - 1000 nm (I'd prefer to use one below 900)
    That frequency range is a must, though. It's the range in which hemoglobin is only a weak absorber of the infrared light.
    • Voltage: 3.3 V - 5 V (To work with common microcontrollers)
    • Power:  200-300 microWatts as a lower limit for steady light output. I'll say something around 10-20 milliwatts for an upper bound on power
    500 microWatts was reported as the "power" of the diode used in the experiment. Whether that was the wattage drawn by the LED during operation, or the energy delivered by the light emitted by said LED, I'm unsure.
    • Amperage:  Something in the milliWatt range, that's easy to drive and output accurately.
    • Lifetime:  Not really a crucial stat at this time, but obviously something I can use more than 100 times. I'll probably have it running a pulse for no more than 2 hours.
    • View: I'd prefer something that's going to give me a nice, tight beam, but can be modified with a lens to cover a wider area later.
    • Mounting style: Through Hole
    • Size:  Something reasonable. Millimeter scale is preferable.
    • Quantity:  No more than 50. I'll probably end up using more than 1 LED, but I have no idea what I could do with a bunch of low-power infrared LEDs.
    • Style:  Single LEDs. Strips of diodes will just be a pain to work with. 
    • Price Range: Since I'm assuming I'll end up having to bulk order, no more than $5 per diode.

    On a slightly different note, are there any reasonably sized alternatives to Solid State lasers that can provide microsecond or nanosecond speed pulses with a fairly high amount of power (300-500 milliwatt range)? Something with a collimated beam, or a beam that lends itself to being collimated would help as well.
  • ThomasEgiThomasEgi February 2016
    TheGreyKnight, i'm on irc pretty much 24/7, worst case you catch me sleeping.
    You'd have to connect to the freenode irc network and enter #biohack. That's all. I suggest using the same nick as on the forum.

    As for the diodes.
    Best thing that comes to my mind would be the SFH 4550.
    It's rated radiation flux for constant output is 50mW if I remember correctly. 850nm. Switching times 10/90 and 90/10 is rated 12ns. If desired it can be pulsed as hell (up to 1.7A if you keep the duty cycle low). Since it's an LED it scales pretty well into the lower-power range. Standard 5mm LED size, sells for about 0.65€ in single quantities, half the price when you order in bulk of 25 or 100.  It's not a laser-beam you get but 3° half-angle is about as narrow as an LED beam can get.
    As for driving it into ns-pulses I'd recommend to wire up some 74AHC logic in parallel. Following the usual rules for high speed designs.  Might help to bypass the diode's current limiting resistor with a small capacitor to clear charge carriers a bit faster out of the diode.

    As for matching detectors:
    Given the wavelength you may want to use the SFH203-FA variant for receiving (if you don't want to pick up light below 800nm anyway)
    BPW 34 would offer you more convenient mounting (it's flat-top) and a quite large sensor area, response time is still at speedy 20ns. Available in both,SMD and THT.
    Talking about flat tops SFH203-P offers that ,too. Same price ballpark as the other diodes mentioned.
    Getting a signal out of the diodes at such high speeds probably requires more than just 3.3V or 5V. Circuit efforts depends a lot on what you really need to record. Be prepared to use around 30V if you really need to measure those 2digit nanosecond pulses like there's no tomorrow.
  • TheGreyKnightTheGreyKnight February 2016
    Odd. I've connected to that channel a number of times and seen nobody on. I'm going to try bypassing my firewall, and see if that's what's causing trouble.
  • ThomasEgiThomasEgi February 2016
    oh i'm horribly sorry. the channel is ##biohack , in accordance with the freenode channel naming rules. silly little mistake of mine to not notice. See you there :)