#448 – An Interview with Jean Rintoul

Chris Gammell: This is the Ambar Podcast. Released June 23rd, 2019. Episode 448. An interview with Gene Rintoul. Welcome to the Amp Hour. I'm Chris Gammell of Contextual Electronics.

Gene: And I'm Gene from MindsEye Biomedical.

Chris Gammell: Welcome, Gene. How are you doing?

Gene: I'm pretty good. Yeah.

Chris Gammell: I'm excited to talk to you about the intersection of bio and electronics. I think that is like, obviously, it's a way of the future, but it's such an important field. Anything in medical is obviously very important. How did you get into that? Or maybe a little bit first about what MindsEye is meant for, and then we'll get back into what you got there, how you got there.

Gene: Yeah, sure. So MindsEye Biomedical is my product development consulting company. It also holds the Spectra open source electrical impedance tomography kit, which is a low cost way to do biomedical imaging. And I mean, I came to that. I've actually been in biosensor wearables for about over 10 years, maybe 12 or 13 years in startups in Silicon Valley. And I've made a few medical devices as part of that as well. And I sort of thought, hmm, there's some problems with getting, as you said, bioelectronics is really important, but it's really hard to actually get things through the regulatory process, which things should go through. But it stifles innovation. And I was thinking, oh, you know what's a much faster and interesting route to market is to encourage innovation. Because I think the truth of the matter is that it's actually not that hard to innovate in this area. I'm not saying it's super easy and you'll just like fall over and Bob's your uncle.

Chris Gammell: It's not like a content area that's super easy. You're saying it's the area's vast. It's a plum for new ideas.

Gene: It is really plum, but all the people, like so many people are really put off going anywhere near it.

Chris Gammell: Oh, I'm terrified personally. Like, yeah, I just, I don't, I don't even know where to start.

Gene: Yeah. And I think, yeah, so basically you get all these electronic hackers, you've got the Raspberry Pi situation, you've got the Arduinos and the maker movements, although, you know, that's obviously having a little hiccup right now. But you have all these like wonderful open source movements. And they're just firing along and they're teaching people. More people are learning how to, you know, get involved with this sort of hardware, software knowledge base. And that's really awesome. But I haven't really seen the same thing for biomedical. Right. And that's because it seems really threatening. And so I was thinking, is there a way to get into it in a really safe way? So you want to be safe. I'll just put that one out there. You also can't be diagnostic. Otherwise, or you can't mention anything diagnostic or say anything diagnostic. Uh-huh. As that also means that you should have regulatory approval. But let's just say you wanted to do something like see inside the body. Sure. Now, yeah.

Chris Gammell: You got a tummy ache and you're like, what's going on here?

Gene: Did I swallow some nails or like, you know what? Yeah. Right. So, and I think that's like seeing inside the body is actually a really good place to start. So if you can do that in a safe way and get information about what's inside that you can't see using your own eyes, that's really amazing information.

Chris Gammell: Right.

Gene: You could put it on your, you know, measure your stomach changes, as you just mentioned, or other organ changes, heart changes, lung changes. And we can sort of go into like the whole variety of other applications. Sure. Later.

Chris Gammell: But yeah, well, I was, I was thinking too of like, when I, when I think about like looking inside the body, I think about x-rays and CT scans and like really big, expensive and dangerous things like you're saying. And so I, I guess I never knew this was an option. Like what is, what is, how do you make it so that it actually is available to people?

Gene: Uh, yeah. So that's kind of what inspired me. Um, it's because, so cat scans, um, use x-rays and x-rays are really bad for you. They're actually the biggest source of radiation, um, in the United States. So you really don't want to get a cat. Yeah. You don't want to get a cat scan. Don't get a cat scan unless, you know, you, you know, you really, really need one.

Chris Gammell: No casual, no casual Friday night cat scans you're saying.

Gene: Yeah. Yeah. So, so, so don't just, um, you know, you can't really have one in your house. They're a little large. They cost a few million dollars. Um, so it's sort of out of range of the kind of experimental or hacker community. Um, they are cool though. Um, and MRIs are even cooler. Like that is just such beautiful physics involved in an MRI. Yeah. But, um, you know, a few million dollars, you've got to have your own helium quenching chamber. Um, you've got to have dedicated staff that are highly trained to manage it. It's, you know, so two, currently two thirds of the world, like the, the bottom, um, the economically disadvantaged people, they don't have any access to any kind of medical imaging at all. So it was sort of like, well, there's these excellent sort of high end things that we have and there's nothing else. Right.

Chris Gammell: And then the next one down is like eyeballs and like, uh, maybe listening with your ears, I guess, is kind of looking inside with your ears, I guess. Right. Or like a stethoscope, but that's barely it.

Gene: No, exactly. Well, well, ultrasound is like, you know, the, the, the, the ear extension.

Chris Gammell: Well, yeah. So, um, past, past guests of the show, uh, Greg Charvat has been, he worked on the butterfly. That's the portable one, but even that that's still $2,000, right? That's still pretty expensive and based on smartphones and everything else. So.

Gene: Oh, yeah. Um, and I don't believe it's open source. That's true. That's really sexy. I believe that. Is that the, the memes based ultrasound?

Chris Gammell: That's right. Yep. Yep.

Gene: I love that. Um, and I think that's super awesome. And, you know, I want one. Um, and, um, I would love to be playing with a kit. Um, I would love to make a kit with memes based ultrasound, like ultrasound sensors, but to tell you the truth, I've actually had to look for them. I can't like just buy them. I can't, you can't readily buy those just yet. Cause they're all super patented into these. Uh, like there's a few different companies doing them and a few different labs that are sort of like working on them and they're, they're awesome. Um, but I can't, I haven't been able to get my hands on one.

Chris Gammell: Yeah. I think it's, that's probably the, the, the most secret of the sauce that they have. So I'm sure they're keeping that pretty close right now.

Gene: So there's huge potential there to unlock this area. Um, as I said before, and what I did is look at, um, there's a whole bunch of labs that are using current based techniques and why that's kind of cool is, um, because, uh, you don't need any special equipment. You don't need a specially manufactured, you know, wafer scale MIMS process to, to make your special ultrasound sensor here. Um, you just need AC current. Yeah. So you need alternating current. Um, and that's it. And I'm using alternating current between about a hundred Hertz and 80 kilohertz, um, to get different information about the body. And then it's actually using the same, uh, reconstruction technique that a cat scan uses, uh, which is an inverse radon transform to recreate an image. Now the image itself isn't as high resolution as an MRI or a cat scan, but nonetheless, I'm getting an image and, um, you can get an image with AC current, which I think is actually kind of cool.

Chris Gammell: Can, can you walk us through the, that's, that's very cool. Uh, can you walk us through like the physics of that? Like how that works? I, I, I don't quite get that.

Gene: Well, yeah. So, uh, so let's just like think of the body as, um, a circuit, um, for a second. And I think that's the like key is just imagining, um, a single cell. Um, and if you look at a cell, you can, um, so you've got extracellular and intracellular fluid, and then you have the cell membrane. The cell membrane is kind of like a capacitive layer. Um, and, and the different intracellular extracellular fluids, they have different resistances. So you can kind of model a circuit right there. Now you have a whole bunch of these cells stuck together and different cells from different parts of your body are going to have, um, different, different values, different resistances and capacitive, capacicences. So, um, basically what that gives you is a different dielectric spectrum when you send different alternating currents through your body. So an example of that would be if you send a, um, like, uh, a high frequency current through, it will go straight through cells, getting the intracellular fluid. If you send a lower frequency current through, say like around, um, one kilohertz, it will get the extracellular fluid instead. So using that information, you can differentiate different materials in different locations. Um, and the basics of that, that's the basic methodology underneath. On top of that, you can apply all these mathematical algorithms, but the basic understanding is just that, you know, the body is this network of resistors and capacitors. Um, and we're just modeling that. And then there's a few ways that you can solve for that. One of them is solving Maxwell's equations. Um, but there's a few other techniques that we can go into as well. And the base tomographic technique is just like, so you send current, um, in, and then you measure the voltage on the other electrode. So say you have a tank with 32 electrodes and you send, uh, uh, a signal in and you measure it on the other electrodes. It will be perturbed based on what or impeded would be the correct term because it's impedance tomography, uh, uh, based on, uh, what's in the way. So if you have a lung, if you have a piece of bone, maybe you've got some muscle or some fat or, I don't know, brain or something, um, bits and pieces of conductive stuff, all with different, uh, uh, cell circuits, um, you can model those different paths and do this reconstruction by sending through, basically you're sending through these currents at all these different angles. And then by, uh, doing inverse radon transform, you can then reconstruct an image.

Chris Gammell: Hmm. And how, what, what's like the levels of current level of current that's going through? Cause I, I, I imagine like electrocution, obviously that's not the case here, but like, what are, what are like the relative levels of current we're talking about?

Gene: Yeah. Yeah. Um, so it's in the, um, about 10 microamp range. Um, so if you want to put that in perspective, um, say you might've heard of TDCS, which is transplanial direct current stimulation, there's also TACS, which is just alternating current version of that, that uses milliamps and that's barely perceptible, but somewhat perceptible. Uh, you can feel that, um, this is an order of magnitude below that. And, um, if you take EEG or electroencephal, electroencephalography, um, that's, um, a very well known technique, uh, to measure the voltages, um, passively, uh, on the, you know, in neuroscience, um, that actually measures contact impedance by sending a small current through as well, which is an order of magnitude below, uh, what I'm sending through an EIT. And the level of current that's being sent through is actually, uh, within the IEC 60601-1, uh, guidelines for safe use on humans, um, as well. Um, so basically, yeah, um, that's kind of cool because, um, because it's a safe technique. There's no x-rays involved. Um, the current that's being sent through is super low, um, so low that it's not going to interfere with any, say, electrophysiological, uh, electrophysiological process. Um, so that kind of makes it like an open platform to experiment with, um, and sort of do data science on if you want to do machine learning on the data.

Chris Gammell: Right. Cause you start to then put like, put random patterns in and just see what happens. Or if you already know what a certain structure is inside of the chamber, the testing area, you were saying you kind of back calculate that out with machine learning over time.

Gene: Um, not with, oh, not even with machine learning, just with a base, um, tomographic reconstruction algorithm. Um, you can get a, like an image reconstruction and then you could do machine learning over that again to get even higher definition information. Okay. Um, so it, the other thing is, um, most of the reconstructions only use a single frequency, usually around about 50 kilohertz. Um, you can also do multi-frequency electrical impedance tomography where you send, um, a whole range of different frequencies through, and that will give you the dielectric spectrum at each, uh, location in space, um, which gives you like a huge amount of information about what material is at any location in space, but it's so high dimensional. It's very hard to recreate an image from that. Um, but it's, it's, uh, it's all unique information, um, which is kind of cool. Cause so the, the most recent results that are coming out in papers on this technique, they apply deep learning to a reconstructed image. They do use, uh, like a reconstruction algorithm first to, to get a basic image and then they improve the image using deep learning to like sharpen it up and they can actually really increase the spatial resolution that way.

Chris Gammell: I'm looking at some of the, so you had sent me a presentation and I, uh, hopefully we can link this or at least you had done talks to, we'll obviously link to every, you know, all your talks and stuff like that. I'll probably ask you more about those too, but, um, I'm looking at some of the, the, the, the reconstructions that you're talking about and I'm trying to figure out like the, the 2d versus 3d type of thing. Is it like, it seems like the images are 2d right now, but is it just, there's simply a matter of passing it and regular intervals over a sample in order to get, to get more resolution, to get a third dimension rather. Like how does, how does this work in two and three dimensions? I guess.

Gene: Uh, yeah, yeah, yeah. No, that's a good question. Um, so I've only done it and I've just been doing it in two dimensions. Um, and, uh, so that's the easiest way to do it. Um, and you can absolutely do it in three dimensions. Um, you just need to have an electrode array that also covers those three dimensions. Um, so one way to do that would be to have like say multiple, uh, cylindrical, uh, arrays in a tank. Another way to do that. Well, so, so it doesn't just have to be a tank. You can also wrap it around your body, um, and do it that way as well. You just have to wrap it around something conductive.

Chris Gammell: Which the human body is pretty decent at, I hear. Yeah.

Gene: Yeah. Yeah. 80% water.

Chris Gammell: Yeah. Salty, salty too.

Gene: Yeah. Yeah. So, um, yeah, you can absolutely do like 3d reconstructions or 2d reconstructions and, you know, it's up to, you know, people to sort of do that themselves on the thing they want to reconstruct. Um, but I've given 2d, uh, examples of it.

Chris Gammell: Sure.

Gene: Um, but yeah.

Chris Gammell: That's good. I think for like visual, like you're, you're doing the, the cross-up I think too, to, to get the tool out there and I believe build a community around it. Um, and that really kind of helps to bring people in, I'm sure.

Gene: Yeah. So that, the idea is, so this technique has only been used in labs so far. Um, and there's actually, there is an effort that's sort of hitting some hospitals in Europe because it's easier to get things out in Europe for, uh, lung imaging, um, and also, uh, sort of, uh, and also monitoring breathing and, uh, premature babies. Um, so there's, there, oh, and there's another, uh, interesting tool that has come out to, uh, be able to non-invasively diagnose cervical cancer. There's a company that's, um, so all these women who have pap smears and things, um, and that, you know, they, they scrape something out and then they send it away to a lab. And then like, you know, a few weeks later you get a result and the whole thing was kind of unpleasant. Um, what if they could just like stick this thing in and they don't need to do any sort of, you know, tissue removal or anything horrible like that. It just goes beep and it tells you, you know, good or bad. Um, and that company is called Zillico and that's using a spectroscopy, um, which is, um, like the simple, just, um, uh, two electrode version of this technique, which is sort of like, Oh, that's, that's a great use for it. So they just, they're, they're just sending multiple frequencies through and getting a dielectric spectrum.

Chris Gammell: Yeah. Um, it's really interesting to hearing, kind of hearing how this is like a burgeoning, uh, like it's not, so you're like, like you said at the beginning, it's not diagnostic. You're, you're very careful about that too, because it's, it's not regulated or anything like that. And yet it's, it's used as a tool for potentially like as, as researchers and stuff like that. So where, where is, where is that line?

Gene: Well, I'm not making a claim. So, um, this is a piece of open source hardware. Um, so, and it makes no claim. Whereas the company Zillico, um, that's not an invasively detecting cervical cancer. They are making a claim to detect cervical cancer. Yeah.

Chris Gammell: And so they have to, that's a big claim too, like quite big. I'm sure people would be very upset if that was right or wrong in the, in the wrong direction.

Gene: Right. Right. So a lot, and it's good that it's, um, you know, tightly regulated and they have to have a really accurate result. Um, and they're using, um, like the, the simplest application of this technique to do that. And, um, so although Spectra is doing exactly the same thing, like the technique is the same. It's not saying it can diagnose cervical cancer. They've also got to like, mind you, they would also have all the data from clinical trials and they would have done, made an algorithm over that data to try to, you know, up their accuracy. So they've, you know, they, they created a, a bio-impatant spectroscopy device. Great. And then they also did like a huge amount of work and, you know, kudos to them gathering all that data and analyzing it. Um, and then, you know, putting that through the FDA, FDA or CE mark, and then actually, um, being able to diagnose. So the technique underneath that is really accessible and it's really powerful and it can be, it can't like, you could do other things with it too. Um, so it's not just cervical cancer limited. And I think that's part of the, what I want to get out. So like, you know, there's other applications in, uh, tumor detection, um, particularly like good at soft tissue tumor detection. And so there's like a whole lot of papers on that, um, like breast cancer and things like that. Um, there's, um, uh, oh, gestural control. Um, so yeah, like people with prosthetic.

Chris Gammell: So like a severed limb or something like that.

Gene: Yeah. Yeah. Like people that have their, uh, limbs amputated, um, you know, um, they, they get these little, uh, they get those extensions, um, and they, you want to know, uh, what their, the muscles in their arm are doing so that you can try to recreate that with the robotic prosthetic limb. Um, so this has a lot of interesting applications there as well, um, because it actually gets an image, um, of, of how the muscles are changing in the arm. So, yeah. Um, so there's a lot of different sort of applications, which are sort of interesting. There's, and they go all scales. There's like lab on chip applications as well, um, which is like super tiny. So, uh, it's very scalable. It's the image resolution is kind of dependent on the, uh, spacing between the electrodes. So if you have really teeny tiny electrodes, you can do it in a Petri dish, but they've also done it on a much larger scale as well. Um, and they've actually used it, um, to, uh, in oil fields to identify, uh, oil deposits, uh, which obviously have a different, um, dielectric constant than the surrounding soil. Um, but then to do that, they have these giant stakes, which are the electrodes that they sort of stake into the ground. Um, and then they send relatively large currents through, um, to do that, uh, conductivity impedance reconstruction there. So you can do it on any scale, but it is, the resolution is dependent on electrode spacing.

Chris Gammell: Okay. All right. That's good to know. Yeah. How did you, uh, I mean, how did you come upon this technique and like, why, why doesn't this exist in the world yet? Yeah. At least maybe as a, as a, I guess you're moving into lower cost, open source kind of realm. Maybe it does exist other places, but how did you come upon all this in the first place? Maybe take us back a little bit.

Gene: Uh, yeah. So, uh, I mean, I've been in a few different biosensor wearable companies that are like making watches that have like heart rate and, um, accelerometer on them. And like, I've actually done two of those startups and they're really cool. Um, don't get me wrong, but I, I sort of, while I was sort of thinking about what information you can actually get out of those watches, you know, like the Apple watch now. Um, I don't think it's, there's, there's not a lot of new information coming out. Um, I don't see there being new sensors that are adding onto that. And I think the accelerometer and the PPG center have been sort of done quite well and, you know, definitely done. Um, and we've got the information that we can get out of that. And I was wondering, oh, is there another, another safe way to get bio information out of the body? Um, and I actually sort of came up with a startup idea, um, and I pursued it for a while. Uh, and, and actually I got into an accelerator as well and moved to China. Oh, wow. Um, yeah, it was hacks. Um, yeah. Uh, and it was called Lear and it's a bladder fullness device. So this was my first foray into this technique. So, um, and, uh, in my first iteration, I actually just did it with a phone app using, um, at the time my phone had, uh, um, uh, an analog input port. Otherwise, you know, no one.

Chris Gammell: Oh, headphones. I miss those things. Don't you?

Gene: Right. Right. You know, a headphone jack. Um, and, and, uh, yeah, I, I, I generated like audio, uh, signals and I injected them in and, um, uh, just sine waves. And I sort of, uh, noticed this sort of change and I was going to all these different hackathons, uh, where there was a lot of doctors. What I was looking for was something that people really needed, um, a wearable that people really needed so that they would keep wearing it. Cause I'd noticed a lot of the problems in the wearable world, like people buy it, buy it. And they like a week later, you know, it's like, it's in the desk drawer.

Chris Gammell: Yeah.

Gene: And I was like wondering how to overcome that. And I think, you know, you can pick people with chronic diseases and illnesses, which means unfortunately you have a much longer path to market. And I hadn't like, I'll just be perfectly honest here. I hadn't, uh, fully, um, inspected when I sort of chose this path, exactly what that path to market meant. But in the meantime, I got really fascinated that I could detect bladder changes, um, with, um, the simplest version of this technique, which started, we started with a two electrode version of this technique, which is just, you know, sending, um, you know, uh, a current through and measuring the impedance change over time. You can just like take a whiz and you notice a change. And it was like, isn't that interesting? And you just stick it, you know, on the outside of your body. Um, so sort of between your belly button and the top of your pelvic bone.

Chris Gammell: Really? And so it just would, I mean, just because there's less liquid in there then, and it's just a different, different size capacitor, I guess.

Gene: Yeah. So your bladder, yeah. Your bladder is just, you know, full of conductive liquid and you just need to measure that. Um, so, and, but there's better ways to do it. And so this, this technique was prone to noise. And so then we moved to a four electrode common mode rejection, uh, technique. And then, um, you know, there was still a few problems with sort of movement and things like that, but that wasn't the major problem with this concept. Um, so people want this product. So there you go. There's a product idea. Um, the major problem is actually, um, the, the regulatory pathway on it, which is a really horrible and sad thing to say.

Chris Gammell: Cause that would, you're saying, so now maybe you could talk us through like the difference now, this would be a diagnostic device you're saying.

Gene: Yeah. So I, I was hoping that it wouldn't be initially. And I actually had a conversation about it, um, with, um, some people at the FDA, um, trying to, uh, uh, propose that it was a device to mitigate against embarrassment. It was aimed at people that had a neurogenic disorder. So, so that's like people who've had prostate cancer surgery. Often they like accidentally, uh, cut the nerves and stuff. And, you know, if you can't fill, if your bladder's full, uh, your own backs up into your kidneys and you get UTIs and it's just, you know, a bad scene. Um, people who are paraplegics and things like that. Um, so yeah, so basically I sort of, um, they realized that it's not just a device to mitigate against embarrassment. Um, even though it is within all of the biofeedback guidelines, it's actually a de novo application for a medical condition. Cause the only people that would be interested in buying it are people that have a medical condition.

Chris Gammell: And can, could you explain what de novo is? Is that a, is that a term that the FDA uses?

Gene: Yes. Yeah. Um, it means it's new. So if you've got a device that can piggyback on something else, that's already been approved by the FDA, um, you can get it through much faster.

Chris Gammell: Oh, there was the, um, John, John Oliver just did a big thing about that. Um, cause he was talking about, it was about implants. He just did one about, um, hip, hip replacement implants and like kind of how the, the stuff that piggybacks sometimes is goes way too far and does really bad things and they don't get enough approved.

Gene: I think I might've seen something like, uh, a documentary on the hip implants at some point. Yeah. And, and, uh, they were like leaking chromium or something into the body and they had to be removed. Yeah.

Chris Gammell: It was something bad. It was some real bad. Like the metal. Yeah. I think it was the metal based stuff. Oh, they, uh, they, they were grinding, I think cause there's a ball and socket and then the ball was grinding into the socket and then it was like head shavings that were going off or something like that. It was, it, yeah. Gave me the, gave me the willies.

Gene: Yeah. Yeah. So, yeah. So it's sort of interesting what you like, where, like it seems very uneven and, you know, things, things, you know, different approaches can, can change this. Um, and I sort of thought about it and, um, I decided that, um, it was actually going to be a really long path to get to market. Um, and investors, some, some investors love medical devices, um, and many don't because they're really painful, particularly in the United States, unfortunately to get to market. And I had a few other things going on as well at the time that was sort of, I was like, well, now's not a good time for that. Um, so most investors have like, what is it? 10 year funds or something. So they have to get a return on their investment. And by the time you probably get some money, they're probably a couple of years into their fund or something. Um, they've been scouting for a little while. Um, so you have to go through all that in a short period of time. So I sort of thought about that. Um, and I think, um, I would be prepared to do that again for the right thing. Um, but there's, there's other problem is that I don't actually have a neurogenic bladder.

Chris Gammell: Like a, like a test case that you can use, you mean?

Gene: Oh no, no. I can definitely find test cases and everybody has a bladder and urinates, which is just great. So you can, that's great. Um, um, so yeah, it sort of matters how personally.

Chris Gammell: Our market size is 7.5 billion people.

Gene: That's right. Everybody pees sometimes. Um, uh, yeah. So, yeah. So basically, yeah, it seems that the bigger problem, um, is actually that you can come up with these ideas and I think some of them are good and some of them are bad and like how, how it's good that somebody is filtering them. Um, but there's all these things that you can do technologically, um, in the field of say bioelectronics where all wet electronics in the body meet, um, that seem to be completely unexplored. Um, and there's a lot of low hanging fruit. Um, and, and the question is, how do you, how do you access that or inspire people to play with it? So on the flip side, you know, like let's just forget about the difficulties and like think about, you know, human evolution for a second. And, you know, if like I know in Silicon Valley here, people are talking about, you know, artificial general intelligence and how machines are going to take over and that's all well and good. Um, but I think the only way to really compete here is for us humans to change too. Um, we need to kind of join with machines and, um, I guess that's the, the transhuman view. Um, and, and we need to do it fast. We need to get a move on. Um, it needs this area of technology. Joining with machines. Initially it will improve our healthcare or personal personalized health and longevity. Um, but longterm, um, hopefully it does a lot more than that and makes us, um, even not, not just healthy, but even better people than we could have ever imagined being. Um, right.

Chris Gammell: And I feel like whenever transhumanism comes up to, it's like, that's like a scary term in my head, but it's, I always like scale it back from like, it's not like, it's not cyborgs or sorry, what's the one in Star Trek? Um, it's not the board. Yeah. It's not the board necessarily where you have like, you know, you're half human, half machine, but you're not, I'm not going to say what you're thinking, but when I think about it, it's more like it's monitoring internally. It's having, it's having like insulin pumps when you need them. It's stuff like that where you have, there's interfaces to the body, but not necessarily like replacing all of the body. Is that how you view it?

Gene: Yeah. So I think we can, um, I mean, partly, I think that's where it will start. We should have, um, there's an open source insulin pump project too. That's pretty cool. And has taken a really interesting approach. I should, um, I forget their name right now, but I should link you to, to them later. I've spoken with them. Um, um, yeah, the way that they got around it, cause their thing kind of is diagnostic is that they've just put the plans online.

Chris Gammell: Yeah.

Gene: Um, and you've got to make one yourself and, and, and so you take all the risk on, but then your blood sugar. And if you're diabetic, that's obviously, you know, a life or death situation. Right. Yeah. It's very, very serious risk you're taking.

Chris Gammell: Um, it sounds like in the same manner too, you're, you're kind of saying like, okay, say you are going to take these risks. You don't even have the way to peer into the body. And that's kind of like the tools and the things that you're, you're working on here with spectra. You're giving more insight into the body, just like a sensor might.

Gene: Yeah. Yeah. I mean, it is a sensor. Um, and, um, and by looking into the body, you can determine all sorts of things like a low hanging fruit for this one might be pulmonary edema or water on the lungs. And there's some initial really great research on that already, basically showing that you can do it. Um, and, um, so there's a problem. How about looking for mucus nodules on the lungs that happens to be tuberculosis and cystic fibrosis? Um, cool. Those are, there's another two things. Um, what about, um, you know, quite a few people bought them for sort of interesting applications. Somebody is looking at trying to detect cavities and teeth with it.

Chris Gammell: Oh, wow. Nice.

Gene: Yeah. Um, so, um, so there's all these sort of different things. So, so if you imagine sort of imaging costing a few million dollars, um, and it's a general tool, an MRI or a CAT scan, it can be applied to all sorts of different things. Um, you know, bone breaks, wonderful, um, or not wonderful, but, you know, it's great that we can see where they are.

Chris Gammell: Good to know. Right. Right.

Gene: Yeah. And then we can see how bad it is and then we know what to do. You know, awareness is the first step to doing the right, like making a good action.

Chris Gammell: Right. Right. Right. If you can't, if you can't measure it, it doesn't exist kind of thing. Right. But in this case, you might feel it, but it doesn't exist. That bone, that bone is not broken. It just doesn't exist.

Gene: Exactly. So, yeah. So, um, if we can create like, but the problem here is that, you know, it's actually hard to get this information out. It's not readily available. Um, as I mentioned before, two thirds of the world doesn't have any access at all to medical imaging. Um, an example of where, um, we, you know, should be using MRIs all the time, but don't in the United States is in, um, breast, uh, breast cancer imaging, which I happen, think happens to women over a certain age every year. Um, they, they sort of do this like, oh, do you have any tumors in your breast or something test? And, um, the best way to do that is with an MRI, but we don't do it with an MRI because the costs are inhibited like, like too, too high. And the, the weird thing is why hasn't, why is an MRI still so expensive? It's been around for a really long time. Right. And it's like, it's not coming down in price because it doesn't need to.

Chris Gammell: I always assumed it's because there's like huge magnets in there too. I figured that the, but like you're saying, yeah, I guess there's no, there's not a lot of new players in the market, right? Because they can't get the same dollars that a GE medical could or whoever else is Phillips or whoever makes them these days.

Gene: Yeah. I mean, I think that's part of it. And I, I think a lot of people just like see it as like not an area, like, like to innovate in. So there's like a whole bunch of people that are into ham radio and that's super awesome. And there's, I don't know, like a Lyme SDR and like, there's all these people that are into software to find radio. And that's like a whole movement and all these people are excited about it. And so it's moving forward. But I think we could do that with biomedical imaging and we can do it safely. And as long as you're doing it safely and you're not hurting anyone and we're not going to talk about diagnosis too much. Um, we're just going to like look inside and, um, we can notice like muscle movements or sort of nerve, like, uh, peripheral nerve activity and things like that. That seems like a, you know, you know, no harm done.

Chris Gammell: Yeah. I think, I think my brain at first is like, well, no, I can't, I can't look at my, you know, like I can't do that. That's what doctors do or that's, you know, and I think that's like a, it's kind of baked into my brain at least as well. I'm just speaking personally. Um, yeah, but if you, like you're saying, like if you, if you make a lot of things, if you start to make this tool and you make it more accessible, then people get more comfortable doing these things. And yeah, I mean, I think that that could be something that opens up new innovation paths for, I guess you were talking about even gesture control, right? It might not necessarily be in a medical context, but it could be the same technology that gets used in because of the, because of the things that it enables.

Gene: Exactly. Exactly. So it's like a general technology. Um, it's a general imaging technology and it can definitely be improved on and, um, and it can enable all sorts of different things in these different directions. And, um, just the idea or putting that idea into people's heads that this is a hackable area. It should be like raspberry pies and Arduinos. Um, I think it's like, yeah, just look at that. You can also just look inside your body. Why not? Um, or, or, or, or like the other conductive thing or, you know, that mouse over there or, um, or your dog or, um, all of this is still for long enough for sure.

Chris Gammell: Yeah.

Gene: Right. Yeah.

Chris Gammell: Um, well, let's talk about, let's talk about the, the actual hardware itself. Cause I'd love to kind of like paint a picture for people that haven't seen it yet. Obviously we're going to link to the crowd supply campaign and the images you have online. Um, but if you, if you could maybe paint a picture of what it looks like and then kind of walk us through the block diagram of how it works, that'd be great.

Gene: Oh, sure. Yeah. So, um, basically, um, yeah, um, it has, um, I rely heavily on, uh, analog devices parts, um, cause they're high precision. Um, so what it has is a, like a main MCU that has a dedicated analog front end, um, that does the impedance measurement. Um, it's got like a 16 bit, um, E to D, uh, on it. Um, and, um, the way that, um, that, that, that works is that basically it's a controlled current source. And then I measure a voltage on the return side. So, um, yeah, so basically you have this current and then you measure a voltage on the other side of whatever the conductive object is that you're measuring. Um, I, I like the way that I'm sort of showing it to people is in a tank. So you can reconstruct anything in a tank. You can also wrap it around things. Um, but you know, the tank is really, um, actually pretty awesome because, uh, what the tank gives you is a known geometry.

Chris Gammell: And so when you say tank, what do you, uh, can you explain what the tank looks like too?

Gene: Oh yeah. It's like a, you know, little, uh, it's a, it's a small plastic, uh, transparent tank that has a ring of electrodes around the, the edge of it that are all evenly spaced. Um, and then that ring connects into the main PCB, um, which has this.

Chris Gammell: So it looks like a flex PCB coming around. So like around the edges, flex PCB or what does that? Yeah.

Gene: Yeah. Yeah. So flex PCB. So yeah. So basically, yeah, the electrode cable, I thought about doing it a few different ways and, you know, for, for some applications you want like separate, separated, uh, electrodes in the sort of way you've probably think of, uh, EEG caps and things like that. Um, but for others, it's fine just to have it in a flex electrode array. In fact, it's kind of awesome because then you can control the spacing between the electrodes, which can help you with image reconstruction because you just need that, uh, geometry knowledge to be able to do, uh, make the best possible image reconstruction. Um, yeah, so basically that main MCU patches out to, um, a bunch of multiplexes, um, which, you know, multiplex in 32 different ways that are going out to the electrodes. Um, those multiplexes are connected back to the main MCU, uh, GPIOs so that you can, uh, control them and, uh, you know, control which one's reading and writing at different times. And those multiplexes then connect to the 32 electrodes, which are in the tank.

Chris Gammell: So, so would you select, so, okay. So, and I, it kind of looks like this is going to sound silly. It kind of reminds me like the bullet time rigs. Have you ever seen those things from like the matrix and like, you know, like it's circular and you're going to be sending stuff across. You're kind of like doing anyways. Um, like you're, you're basically, no, no, uh, I haven't, I haven't, I mean, I've seen

Gene: the matrix, but I haven't seen the way they, yeah, they did it with.

Chris Gammell: And, and it was reconstructed by, um, uh, the crontech folks. They do like high speed cameras too. It's basically just like, it's similar to this where there's, it's on a ring and there's, there's multiple points around it all equidistant. And so the idea being that you can, if you're turning one thing on at a time, you get like a view around this centralized thing. So I'm looking at the picture you have of the tank filled with broccoli. And so the broccoli, you would get like the electrode kind of like would be activated around each, around 32 times around the broccoli. Right.

Gene: Exactly. Yeah. Um, yeah. So I've actually got these like, um, like a few articles actually. So you don't see all of this stuff on open EIT.github.io and you can like see the reconstruction patterns. Um, so you can actually like pick different reconstruction patterns as well, uh, with spectra. So you could try using just eight electrodes and then move to like 16 and move to 32. So the more electrodes you have, um, the better the spatial resolution you get. And I think it's kind of interesting to see, you know, what the difference is between eight and 16 electrodes. Um, cause then you go, Oh, I add more electrode and, um, the spatial resolution improves. Um, and then, yeah. And then there's different patterns, current patterns that you can send through the different electrodes. Um, I've just picked, and this is all open on, on GitHub. Um, and you can see the stim patterns in the firmware. Um, and you select between them using like a dashboard. So there's like a, um, a dashboard that you install on your computer. Um, and that lets you control the firmware. Um, and you can like, you select different algorithms, different, um, stim patterns and things like that. And you can sort of see, see these differences in the spatial resolution of the reconstruction of the thing that you stick in the tank. Maybe it's a cup, maybe it's a broccoli, maybe it's a fish, um, whatever, whatever you want. Um, maybe you wrap it around your wrist.

Chris Gammell: You were saying stim pattern, like stimulation patterns, is that right?

Gene: Yeah. Yeah. Yeah. So, so, um, yeah. So basically by stim pattern, I mean, um, if you've got eight electrodes, there's a whole bunch of different ways that you can sort of, you send out a signal and then you read it on the other electrodes. Um, and there's sort of like what ordering you do that in, um, and 16 electrodes has a different stimulation pattern again. Um, and you can get different results dependent on how you sort of space things out. But basically you want as many different, uh, mixes between the electrodes as possible. So you're trying to get, like, if you imagine like a straight line path between, um, electrodes, um, then, uh, you want as many of them as possible to be able to do a spatial reconstruction. So you'll get, um, typically you'll get like a higher density of those lines around the edge of, um, um, the reconstruction and fewer of them go through the middle, which means that just, if it's a circle, you'll get, um, slightly lower spatial resolution in the middle than you do at the edges.

Chris Gammell: Oh, interesting. Okay.

Gene: Yeah.

Chris Gammell: And then, so, um, so when you talk about the multiplexing as well, so you're pumping current out through, let's maybe just say one through 32, right? So, so you pump it out through electrode one. Does that mean you're using, uh, like 17, like across the way to detect then?

Gene: Um, yes. Um, and also 16 and, um, like all the other, yeah, all the other ones to detect.

Chris Gammell: Um, actually it's not, it's not just the straight across. You basically use one to radiate and then that kind of goes to all the ones across the opposite side you're saying.

Gene: Yep. Yep, exactly.

Chris Gammell: So then how do you, how do you actually then sample? So like, what is the sampling pattern for doing that? Cause I'm imagining like an AC waveform. Is it actually AC being pumped out continuously or is it like, uh, different levels being pumped out on a different basis? So like, so if I pump current to electrode one, it's like pump to current current electrode one, measure it 15, 16, 17 or all of them, and then move on and do a different level. Or like, what, what is the, is it actually like a continuous AC waveform, I guess that I'm asking?

Gene: Um, no, well, no, it's not. Um, so it's doing a, like, so as you, as you mentioned, uh, what happens is that you're send something out on say electrode one, you read on all the other electrodes and you get your impedance measurements and then you shift and then you send something out on electrode two. Um, and this is all done with the multiplexes. So they're like fast switching analog multiplexes. So they will, um, you know, um, so for, you know, uh, a very, very split second, um, you know, switch over and stabilize and then you will start the impedance measurement sequence again. Um, yes. So you, uh, so there's different ways to do EIT architectures just to put that out there. This is just one of them. Um, I chose this one, um, because it was inexpensive. So the reason that, so, um, just to give you some sort of insight into other EIT systems. So I didn't invent EIT. It's been around since the nineties, um, in labs and, um, it has taken like a big cart of equipment. So you get your $40,000 EEG, you get your, um, $10,000 current source you get, um, and then you sort of join them all together with an expensive switching system. And, um, that they're like, there are all these like big messes of wires on a giant cart that they like wheel around. And that just seemed so big. And it's also kind of expensive. Um, and I was like, man, you know, I've, I've been doing all this like consumer wearable stuff. So my, my background is making things small. Um, and I was like, you know, I could make this smaller, like, like, cause I was reading about, um, electrical impedance tomography and I became really fascinated by it because it was like, wow, you can image the body, but you need all this really expensive equipment. Wait a second. Do you really? Um, and then I sort of like looked into it. You do need to be able to make precise measurements, precise analog measurements. Yes. But, you know, that's why it's got a few. And like, I mean, they're, I actually like the, the, the most expensive parts of the, the bomb are the, like is the main MCU, which has this like, uh, you know, analog devices, uh, you know, precision measurement system on it. Um, but still that's much, much cheaper than, um, the, the other systems out there. So it's, it's comparatively, um, you know, basically this is a cost effective way to do it by multiplexing the signal around.

Chris Gammell: Yeah. Um, so I was wondering about the, because you had mentioned like an AC signal as well. So, uh, I'm imagining like a squirt of current going into electrode one and then it switches to, to, um, to electrode two or whatever, but is it, is it just a, uh, a short pulse of DC current or is there actually like a waveform that you, that you, um, program to go in there?

Gene: It's absolutely not DC current. Um, this thing actually has some DC filters on it as well to absolutely make sure that that's definitely not the case.

Chris Gammell: Well, I just mean like DC, DC, DC turned on and off real fast is effectively AC. That's what, that's what I'm really trying to figure out too.

Gene: I see what you're saying. So it's got a DDS on it, um, like a, um, uh, digital synthesis to make sine waves. Um, um, so it's making these, um, so you're, you're, uh, and you can actually program it to make any waveform you want as well. I might also add, so it doesn't, you don't just have to do sine waves, but I've done sine waves. Great. Um, you know, sine waves work great. Well documented here. Uh, yeah. And, uh, yeah, so basically it spits out these sine waves. Now it doesn't just like, you know, uh, spit out like a single waveform either. Um, that's not really enough. Um, so at different frequencies you, um, and, and you might want to run this thing at different frequencies to get that different information, which is like really like kind of interesting. Um, you, obviously that takes, um, so it has a, like a, uh, uh, uh, uh, uh, uh, uh, uh, precision crystal on it as well. So basically, um, you've got accurate timing, um, and you, you get, uh, back a magnitude and phase. Um, so impedance is like, what is it? It's like the, the real, real and complex. Uh, so you've got like your resistance and your capacitive element, um, in the phase. Uh, but I'm just using the magnitude, um, of that, of that return amount. Um, and you actually need quite a few waveforms to be able to do a DFT on it. So this, um, main MCCU, what it also does is a real-time, uh, digital Fourier transform. So, um, yeah, yeah, yeah. So say you, I don't know, you send out a 50 kilohertz signal or something. That's a good, why not? Um, then you're measuring that back on the receive and you want to get the magnitude of that. So, um, you do this DFT and you can get that. Um, and you can obviously do that at different frequencies too. Um, and it's very accurate and that's nice because, you know, it's good to have, uh, an accurate reading, particularly for something like, uh, this, where you're looking for these, um, changes and impedance and you really care about what level they're at relative to one another to be able to get any resolution at all. Um, one of the, um, problems with this technique, which I'll just be super open about is when I was talking about spatial reconstruction and this is why it's kind of limited, but it's not necessarily the end of the story. Um, is that, so current doesn't travel in straight lines. So in your tank, um, uh, a lot of the assumptions, uh, for the reconstruction algorithm that the current is going straight from A to B, like from one electrode to whatever the receive electrode is. And that just simply is not the case. Current like bends based on the, uh, the, uh, the impede that the inner impedances of the object. So if you've got a bone or something, if you've got like lungs, which are like air in them, the current will go around them. So it takes all these funny routes through the body, um, which makes it a super hard, uh, uh, makes it a super hard mathematical problem to solve. So people have tried to, um, simplify it in various ways by making kind of assumptions. Like some of the algorithms are like, um, actually like a lot of them, they assume that the body is only a network of resistors and that's okay. Yeah. Yeah. So you're just laughing right there immediately, but that's what they're doing, which is terrible because it isn't, it's got like, it's like a very first order approximation, right? Yes. Yeah.

Speaker ?: Right.

Gene: Yeah. It's so, yeah. So basically there's a lot of open room for improvement. Um, um, because it's not just, you know, a bunch of resistors is like every cell has a capacitive membrane, um, as well. Uh, so, so that's part of it. Um, and so that's part of, part of the problem about how to get better spatial resolution is that, um, there's all these limiting assumptions made. Right. Um, and you know, limiting the assumptions works by the way. It works great. Um, sure.

Chris Gammell: It's a constrained problem set then. And you, you basically put in known things and you're like, yeah, there it is. But then you put something unknown and you're like, Oh, what is this thing?

Gene: Yeah. Exactly.

Chris Gammell: I was going to ask about that too. Cause you had, you had mentioned the different frequencies or sorry, different, uh, different types of cells respond to different frequencies and stuff like that. So does there need to be like a library of like, well, brain tissue responds really well to 47 kilohertz, but a bone responds really well to five kilohertz or like, how do you actually then go about back calculating all that stuff out?

Gene: Um, so, so you can just forward calculate it kind of, or like, um, but that's a really good question. So you can obviously get an even better resolution if you already have some idea about what's in there. Um, but there are actually funny that you mentioned that there are actually libraries of standard and patents data.

Chris Gammell: Oh yeah.

Gene: Um, yeah, yeah. For, for, for all these different bits of your body.

Chris Gammell: Um, it's like black body radiation too. Like whenever that's not black body, but like when you're doing the radiation problems, right? You have to know the material type, but if you're just looking through a thermal camera, it's like, I don't know what I'm looking at. It's just radiating some kind of IR. And it's like, I don't have, you know, you can assume some, some nominal type of material, but if you don't know what it actually is, then you don't get an accurate temperature measurement. And this kind of feels like the same thing in that if you don't know what type of material it is, you don't know really what you're, you know, you know, there's density there, some kind of stuff there, but it's not, not going to be as good as if you know exactly what stuff is in there.

Gene: Right. Exactly. So if you know what stuff is in there, you can do much better again with the spatial reconstructions. Um, but some of the, like, it depends what the question is you're trying to solve. Are you trying to figure out what stuff is in there or are you trying to like, do you already know what stuff is in there and you're just trying to determine its configuration?

Chris Gammell: So it's sort of like, which it's not, is it a bone? It's, is the bone broken or that kind of thing?

Gene: I mean, yeah. Yeah. So, so that might, yeah. So that might change depending on what application you're using it for. And what you're saying now kind of reminds me of, uh, uh, MRI. Um, and, and this is like a blessing and a curse in the MRI world. Um, so I, I used to work at a cognitive neuroscience lab, um, where we had the neuroscience imaging center at our disposal. And I've, like, I've taken MRIs of every part of my body.

Chris Gammell: Um, that's super cool.

Gene: By the way.

Chris Gammell: Um, uh, yeah.

Gene: Um, where was I going with that? Oh yeah. Um, they make, um, so, you know, like, um, often when you're sort of like doing machine learning of MRI data, um, you use these standard models of like, say a brain. And I find, and this is really disturbing, um, because like, you know, my brain is not exactly the same as everybody else's brain, you know, there's right. Because we're not all the same person. And if you sort of try to average across all people, like what are you left with? Because like the devil's kind of in the details sometimes and, you know, we're all beautiful and different and stuff. And that's, uh, part of the question.

Chris Gammell: It's like the, what is it? What did the Marine say? It's like, this is, this is my brain. There are many like it, but this one is mine. Yeah. My brain is my best friend.

Gene: Yeah. So yeah, it's the only one you got. So, uh, yeah. So there's this question about like using like average data over sort of individual data. And, um, there's different routes that you can go with that. If you use average data, um, you can definitely bootstrap a process and maybe get better generalized results, but it's only ever going to get you to a certain success rate. It's not really going to solve the individual's problem. Say the individual has a broken bone or like something that is anomalous, uh, that you want to detect in that person. Um, so if you can sort of use a more generalized approach, um, to like, like a camera is pretty good that way. Um, to, and then, then there's this whole area of image recognition that like is burgeoning and fast moving, um, and, you know, ardently hackable, which is great. Um, so image processing really took off or is still taking off. It's great. Um, and you can do that same thing.

Chris Gammell: Yeah.

Gene: Um, with the kind of sensor data that you're getting from these biomedical imaging devices, um, to just take an example of like Google, I think is doing this. They came out with this awesome paper. Uh, they're doing machine learning over MRI data. I don't know if you've, um, seen, uh, some of the, and the machine, like the, basically their results, um, are competitive with six radiologists diagnoses. So, so it's like one machine learning algorithm versus six radiologists. So it's like, it's, it's getting there, um, what you can infer from this data. So I think I find that really exciting, except for the MRI, the MRI is really expensive and like hard to access. So if we could have a cheap mass producible system and then do the machine learning on that, that would be even better. So people are only starting to do machine learning on EIT data.

Chris Gammell: Yeah. And it's like, it, it's basically more data than feed into that machine, right? I mean, that machine being the machine learning thing or the algorithms that are, are doing the recognition. It's like the limiting factor has been MRIs because they're so expensive and radiologists are obviously an expensive thing as well. You got to train someone for 20 years and they have to understand what they're looking at and whatever. So if you knock out the cost of the radiologist, sorry, Dr. So-and-so, uh, it's like, then you can start to have real impact, but only if you continue to put more data into the system,

Gene: right?

Chris Gammell: Right. If you've knocked off all the cost of the radiologists, you can then really make it a affordable thing. If you get better imaging and more, more of that imaging, like you're trying to do.

Gene: Right. Right. Yeah. So I guess, you know, the question for me is how would you get a new imaging effort off the ground? Like, um, say, say this one, um, because you need data to, to feed into it, right. To make it useful. Um, and then you need people to build confidence in it and you, you need doctors to sort of start using it. Um, and maybe they use it in a, like it's used sort of in a preventative way. Like, um, sorry to use the Theranos example, but I mean, uh, that, you know, like stick it in every Walgreens or something and just do it like a yearly body scan or something. And it's just like, you know, it's not a perfect body scan, but it's like, Oh, warning, warning, go see a doctor about this area of your body. And then the doctor says, Oh, you got a warning. We'll send you to, uh, like an even better imaging thing and have a really close look at that kidney of yours.

Chris Gammell: Right. Exactly. It's like a triaging method of, of like, uh, instead of, and so you wouldn't just go right to the radio, the expert radiologist right away. You'd work your way up through a GP and all the way up to the, to the super specialist with the really expensive equipment.

Gene: Absolutely. Yeah. So, so I don't think it actually has to compete with, um, the top end equipment immediately. It can kind of work its way there. Um, from the sort of the low end, which is also kind of nice in that, you know, it's always nice to help, like, you know, help more people. So, um, yeah, so it can be a low end imaging device that kind of grows from the bottom upwards. And as the techniques and technologies and machine learning get better, becomes more and more competitive with the top end imaging.

Chris Gammell: I wanted to ask a little bit more about the, um, the physics of how this thing's working too. Cause I'm still a little, I'm a little hazy on the, the actual, like, if we can kind of like slow time down and think about like how the signals are moving through the system. I'm kind of curious about that. So say we're using 80 kilohertz, right? Just because that's, you said that's the kind of top of the range. So we're using 80 kilohertz. You pump that to electrode one and then the broccoli gets blasted with 80 kilohertz. Um, and then, uh, how long, like how long is that sample on? Like how long is electrode one on?

Gene: Um, I mean, I would have to calculate that based. So basically you do, um, you also want to average over a few measurements as well to get a more precise signal. So say it takes a couple of hundred measurements and it averages over that. It does the DFT and it gets a magnitude. Um, so you've done a little bit of averaging to get a stable result. And what that result is telling you that's going through the broccoli is, um, basically versus something that's not going through the broccoli, um, tells you where the broccoli is because you've got these different geometric paths that it's taking. Um, yeah. So basically, yeah. So you'll average over, um, uh, actually what's the numbers? I think it's 1032 samples.

Chris Gammell: I just, I assumed it was like, I assumed it was like a fixed amount of stuff. Like, so if it's 80 kilohertz and it's going for 0.1 seconds, you're going to have, you know, however many, however many waveforms it can go into that, however many periods you get in that, in that, that sample of time. Is that not correct though? Is it just like a fixed number of samples?

Gene: In this system, it's actually a fixed number of samples. Okay, great. Yes. But it only goes to a maximum of 80 kilohertz. So if you were to go to a higher frequency or a lower frequency than a hundred hertz, you would need different timing parameters on that. Um, but, um, for, for this, this is a sufficient averaging to, um, get a stable, uh, impedance measurement over that entire range. So, yeah. Okay.

Chris Gammell: Okay. So, okay. So we had the, the, the 10, uh, 1048 or 1032, whatever you said it was. So then that goes through the broccoli. Is it like all of the other sensors are, is it each sensor has its own ADC as well? Or sorry. Yeah. Sorry. Each other sensor has an ADC or are those then also getting multiplexed in as data, data points?

Gene: Um, they're all getting multiplexed. So, um, this is basically just an intricate dance of multiplexes going on. Um, so yeah, you send, receive, make that measurement, um, then multiplexer switch, send, receive, um, make the measurement, uh, multiplexer switch and, you know, wash, rinse, repeat around in your stimulation pattern.

Chris Gammell: All right, cool. So then that gets then pulled back into the ADCs. You then have this data file of however many things that happened there. Um, and then, and then what, then it's like into, you said that's, so the DFT is an internal. Is that right? Like it's already, there's an engine in Silicon.

Gene: Um, yeah, yeah. Um, yeah. So basically that's part of why I picked this particular chip as well as it promising these, you know, high precision impedance measurements at 16 bit resolution. Um, um, so basically you get that impedance measurement of the, uh, analog front end and you can send different commands to the analog front end, but it's like, um, this, um, bit code that you write to be able to access it. And then the main MCU is an arm cortex M3. Um, so once you get it back into, like M3 land, uh, you're, you're relatively golden. Uh, it has all the standard things that arm cortex M3 has. And what it does is it actually sends the data out via UART or Bluetooth. So at this point you don't have the raw data though. You can get the raw data, but what, um, I'm sending out is the, uh, the, the processed impedance magnitudes. So for each, uh, arrangement of electrodes, so each send receive pair, you get, uh, one piece of data and, um, and basically you stream that out over UART, um, or Bluetooth, depending on which way you want to go. Um, and, um, that, uh, is fed into the dashboard, um, which, um, is, uh, is, uh, originally written in Python. Um, but it's, um, compiled using, uh, Electron, which is this, um, cool thing that make, uh, makes you, uh, desktop apps in, in a browser. So basically, yeah. Um, so.

Chris Gammell: I think it's, it's, it's, it's what, it's why Slack is, has so much memory usage.

Gene: It is. It's a completely inefficient. Sorry.

Chris Gammell: I don't know. I mean, if it works, it's fast, who cares, right?

Gene: Um, yeah. Yeah. So if you want to do it the fast way, just, you know, download the Python. So the reason that I did that was because I was also trying to make it accessible. So as some people say, so not, not all doctors want to install like a bunch of Python libraries.

Chris Gammell: Right. Right. You're trying to make it like so that people use it. I mean, that's, that's smart.

Gene: Yeah. So, so basically, um, yeah, you can absolutely do it just, just, and it's got full instructions on how to do that. You know, install these Python libraries, you know, standard ones like NumPy and matplotlib and whatnot, you know, very unsurprising, but you still have to, you know, you still have to have those libraries. Um, or if, if, if you don't, uh, want to do that, which, you know, some people don't, then you can do, uh, install this electron app, which means that you don't even have to have Python on your system because it's using a portable version of kind of encapsulated Python. And as you like made the point, yeah, that's exactly what Slack's doing. Um, and you know, there's a whole bunch of other things out there. And I think this is actually a really cool way to go for like app design in general, if you want to create an app quickly, you can sort of prototype something in Python and then convert it into this browser based app. And I just sort of think, oh, wow, that's like a really interesting idea. It's a pity, as you said, it's not very memory efficient, but maybe, you know, we can improve that.

Chris Gammell: But then it's cross platform then too, which is great for again, accessible. Like, so like as a small team, I imagine that that is just paramount to like, you want to get as many people as possible. You don't want to say like, oh, you have to use, you know, this OS or this thing and this thing, use this library. You're basically making it accessible to not just experience types, but also like OSes and everything else then.

Gene: Yeah. Yeah, exactly. Just like you said. So the, yeah. So initially, you know, um, versioning is such an annoying thing to deal with.

Chris Gammell: Yeah.

Gene: Um, nobody likes to deal with it. And, you know, I'm not like, uh, actually that's a lie. I was about to say, I'm not always running all three operating systems at once, but actually I am through VMware. So I run OS X and then through VMware, I run windows and I do firmware things and, um, EE things on my windows, uh, VM machine. And then I also, um, uh, run Linux as well. Um, again, through VMware. Um, so, so, so basically not everybody does that. And I didn't, I didn't actually do that either until, um, like somebody was telling me that, you know, it doesn't work for them because they were using, you know, an operating system that I hadn't used it on. And I was like, oh, oh, that's annoying. Uh, you probably have a wrong version of something. I don't know what thing you have a wrong version of, but whoops. Right.

Chris Gammell: The best case scenario for like shipping a product would be also to ship them a computer that it is destined to work with. Like vertical integration is really expensive, but damn, if it doesn't work every time, you

Gene: know, like, well, yeah, I mean, you can do, you can do that like with a Docker image or something, I think. So, I mean, yeah, but it's like, again, it's like, I don't know. I just don't find that super sexy for some reason, but it's, it's, it's, it's a totally viable solution. And I've seen some like, uh, like software packages, actually finite element packages. So like, um, to the three different algorithms that come with this one's the basic radon transform and then the two others use finite element models. Um, and for a little while I was getting into some of these like open source finite element, uh, tools and they're so hard to set up and they're so version dependent. And I was in this like version of them start with GNU feels like anything that starts with

Chris Gammell: GNU, just run away screaming. I feel like.

Gene: Yeah. Yeah. Um, and, and one, yeah. So basically they like, one of them had like this Docker image and I, that to help get off the ground. Cause they knew that it just didn't work for anybody unless you had like, you know, this exact version of the operating system and this exact driver and this exact something, something. And it was like, Oh, wow. Yeah. Um, and you know, yeah. So you had to have all like GCC, this one and this one and this one. And you know, um, so yeah, it's really nice to sort of think of ways to escape that. Cause I, for one would prefer not to spend my life in version hell, but I think.

Chris Gammell: Right. I, or even writing software at a certain point. I mean, like you do hardware, you do software, you do it all. Um, and if you're just doing versioning stuff, it's like, Oh boy, it's just, you're just in software land for the rest of your life.

Gene: Yeah. And you're not even in, like, I don't find that very interesting software land either. I would rather be doing something new or not just dealing with versioning issues. Um, so yeah, I, I find that, uh, I mean, unfortunately, like, to be honest, I've spent quite a lot of time dealing with versioning issues. Um, but it's not, not my favorite thing at all. Um, yeah. So, and that was a problem with this project too. It didn't start out in this electron format. Um, I actually, um, started, uh, talking with somebody called Marion. Um, and, um, she actually made, uh, this piece of software called cloud brain, um, which used electron. Um, and she's actually a contributor on the opening it project as well. And she was like, you know, basically telling me about, Oh, version, like if you want to get around it, you could do it this way. And I was like, Oh, that's great. That's so exciting. Um, and, uh, yeah, so basically that's how it sort of switched from its, um, original, it was using matplotlib and things like that. Um, right. And, and, um, uh, what else was it using? It was like using QT or something, um, a QT backend. Um, and then it moved, um, into this, um, server-based, uh, model, which is kind of just runs on a browser, which is sort of more generalized. And it's like, Hmm, that's actually not a bad option when, if people have different operating systems, but there's still some limitations with different operating systems that are actually quite hard to get around. So like for instance, the Bluetooth.

Chris Gammell: Yeah. I was going to ask, what is, what is the Bluetooth being used for?

Gene: To, to get the, the UART data off. So the thinking was that, um, so basically you, if you want a really sensitive measurement, um, you probably don't want a power source plugged in and the, the, the, uh, um, or it connected to a computer and a really low noise power source is actually, so, so you can obviously get these very expensive power sources, which they use in these big carts of equipment EIT systems. But another really low noise power source is a lithium ion battery. That's right. Yeah. Yep. Uh, so a lithium ion battery. Yay. They're, you know, they're all over the place. Great. Let's stick one of those on there. Uh, yeah. And, uh, that means you can't connect to anything else. Right. Um, so Bluetooth. Um, so if you want to

Chris Gammell: get a really clean signal, uh, use the battery. And then, so, uh, it just goes as a serial port then on the computer side as well. It's not like going to a mobile app or anything like that.

Gene: No, no, it just goes to the computer. Um, no mobile app. Cause that's a little bit, um, overly, I mean, it's, this is a open source hardware software project. It's not really a mass consumer product

Chris Gammell: here. Um, right. But it's not, but that's also means that someone could go and make an app for it. And just because that's what open projects are great for, right. That would be awesome. Most welcome to do that. Yeah. Um, that's really great. So you, you are making these right now. I mean, so we, we didn't get a chance to connect when you were actually, the project was going on, but you, you were funded, uh, $37,916 raised on, uh, car supply. Congrats for that. Um, thanks. How, how is the, uh, how's the manufacturing going? Um, good, good. Um,

Gene: so basically, uh, I've been making a test jig, um, using a raspberry pi and open OCD, um, to, to, to flash things and, and, uh, that's all working. Um, I've ordered, so it has some custom cables, um, and I've sort of, sort of, um, organized for them to be made. Um, I've ordered the PCBs, which are the, the ring of electrodes. Um, and I am currently hunting down, um, a couple of pieces on the bomb, um, to be able to get the main manufacturing run started. Um, so there's sort of, that's basically where it's at right now. And then hopefully we can, uh, get that running and, um, then make them all. Um, but yeah, all the parts are ordered except for obviously that the

Chris Gammell: main PCB is the, is the, is the key part. Yeah. The expensive part, huh? And you'd mentioned too. So if any ADI people are out there and you want to support a good project, this is one to, uh, throw some weight behind. This is, uh, those do not, I've looked at, I was just looking at the, the, uh, parts on the, the crowd supply page to the ADUCM 350. We've actually mentioned on the show before, cause I remember when I, when I came out, it came out like four or five years ago, I think. But I just remember like seeing that, yeah, the analog front end stuff was just like, it's really, really good, um, low noise stuff. And then it's got the micro in there too. So it's,

Gene: it's great. I mean, yeah, it's a really nice MCU. Um, it's, it's, uh, I mean, it's, it's low power. Um, it's like precision, um, and it's made for impedance measurements, which is perfect. Um, and it's also using our devices, um, uh, multiplexes as well. Um, the ADG seven, four, four,

Chris Gammell: four of those is not, uh, it's not cheap then, huh? No. Sorry.

Gene: Uh, ADG seven, three, two. Um, yeah. So, so we'd love, um, and, uh, differential amplifier as well from our devices. So it's using, um, you know, I, it's sort of quite a few analog devices parts. Um, and those are the major, major, major, major costs on the bomb there, uh, such as life. But, um, but that being said, they keep all the fancy stuff in silicon these days, you know, that's, that's the problem. Yeah. Yeah. And, and, you know, I can see why, um, I mean, it's sort of, you know, it does mean that you want to buy their MCUs, but if you look at like comparisons, so a comparative current source would be like, um, there's something called the Keith Lee 6 2 2 1. That would be the current source that I would recommend for people that wanted to go even deeper. Um, and, you know, I, I think that one costs about seven, $7,000 just for a current source.

Chris Gammell: So then you actually, uh, I used to work on that. I used to work on the supporting that, that piece of machinery at Keith Lee and I broke a couple of them and it was not good.

Gene: Oh yeah. Oh yeah. That one is commonly used, um, sort of in this area, um, it seems. Um, so yeah. Um, so if you're not sort of using that one, then, um, uh, the ADUC M350 is a good option. Um, uh, but you know, it's, um, yeah, every, everything has its pros and cons. Um, so yeah. Um, yep. So basically, yeah. Um, Cortex M3 is nice as it's, you know, pretty standardized going on there.

Chris Gammell: There's nothing surprising. How is it, how is it programming those parts? I always, I always imagined that the, the ADI programming environment wouldn't have been that friendly,

Gene: but maybe it's not bad. Um, you know, pros and cons, uh, they had some reasonable, um, examples. Um, some things were a little bit obscure, um, but it wasn't too bad to sort of, um, make, um, sort of custom analog front end, um, um, um, sort of, uh, it's, uh, uh, bits of code as well. So, so yeah,

Chris Gammell: I mean, it was fine. Um, but a good starting point and being able to build off that kind of thing.

Gene: Yeah. Yeah. So they have a whole bunch of examples, which are pretty good starting points. And you can kind of, uh, I recommend starting with the starting points, you know, always a good place to start.

Chris Gammell: Do you, so how much do you, how do you see you reconfigure like the multiplex and it happens external to the parts. I'm guessing you have, you, you control that logic that actually turns on the different paths and stuff like that, but then internal to the ADUCM 350, does that have a lot of like configuration of the, the muxing between different, um, inter input lines as well? Or is it more like you plug in inputs one to 32 and it just does its own thing then? Um, so, I mean,

Gene: you know, uh, I'm controlling, like, so, so there's custom code to control all the multiplexes. So it doesn't, it's not made to, so, so, so the ADUCM 350 is not made to be an EIT system. Um, so it's not made to like do that, but you can also make it do that. Um, and, um, that what it's actually made for is these, uh, precision impedance measurements, which is pretty good. Um, and, uh, so there's a lot you can just do with, um, just impedance measurement by itself, which I could also get into, um, as, as a, as a, as another topic. And there's a mode in firmware that you can, uh, you can go into and just look at the time series impedance data, which is super interesting. You can see breathing and heart rate and, um, that's actually like, it's different from an ECG signal. Um, so you can sort of see these material changes through any object, which is actually pretty awesome. Um, so yeah, so basically then once you send the command to the analog front end on the ADUCM 350, um, it just runs that entire command until it exits. Um, so there's, yeah, so there's not a lot of back and forth, um, there, well, I mean, there's, you know, you know, in an,

Chris Gammell: in an out. Right. It's not like low level control. You're saying it's just kind of a subroutine that, that, the, the parts set up to do. That's right. That you, that, that you execute

Gene: and you can change that subroutine. Um, and, uh, as I mentioned, you can generate any waveform you want of it, but I'm just generating sine waves. Um, so you can like, so all of that is configurable. Um, and, uh, and you can see the examples of that. And once you send that, it just runs until it finishes. Um, and, and then you could multiplex and run it again or you, or, uh, do whatever you want. Yeah. Yeah. That's great. That's great. Yeah. Yeah. So, yeah. So basically, yeah. So it's the health, it's a health meter on a chip is actually how that, um, part is advertised as well. Um, which is sort of interesting, um, in a lot of ways that sort of, um, you know, they know exactly what kind of applications that they're going for, which, you know, isn't, isn't that surprising, but yeah, that's great that they, they've figured that out. Yeah. I'm looking at, I'm looking at the data sheet

Chris Gammell: right now. It says point of care diagnostics, body worn devices from honoring vital, vital signs. And then amp, well, I don't know these words, amperometric, voltometric and impidometric measurements. So that's probably what, that's the kind of same thing you're talking about.

Gene: Yeah. Um, yeah. And as I mentioned before, so, um, in, in so far, I'm just using the magnitude of the impedance signal, but there's also a phase. Um, the phase actually has a lots of information in it too that could be used. You know, I just haven't been bothered personally. Yeah. But it's there.

Chris Gammell: Well, and that's like something as you build your community out too, like I imagine like there's that, there's the pulling in the phase information. Like you said, you're using sine waves right now. I'm sure that there's something, if, you know, you start doing different waveform excitation waveforms, you could probably do other things in the future, but you're, you're building like a scaffolding for all this stuff to happen, which is great. How has been, how has the, uh, how has the community response been?

Gene: Um, I mean, it's good. It's, it's sort of tricky as well. Um, so, um, there's a bunch of people who bought them for sort of interesting, different, uh, applications. A lot of them definitely sort of commercial in nature. Um, and you know, there's about like seven people that have sort of contributed to like the open EIT GitHub, uh, uh, repository. So those are people that have sort of contributed to the project. Um, so yeah, so, so basically there's contributors and then, um, I mean, the major problem has been that, you know, not everybody has one yet because hardware problems

Chris Gammell: are still the thing. Atoms are not, not easy to get ahold of sometimes.

Gene: Yeah. So, um, yeah, so there's like, so there's, you know, about seven people contributing. Um, there's a discord channel as well for, for chatting. Um, but not everybody has one yet because I only just finished the crowdfunding campaign. So once everybody gets one, um, that has been interested, um, and starts using it, I think that will be, uh, really interesting to see how they start interacting and what they start doing. So, um, I, I, I, I try to sort of like suggest, um, like you can, you can put issues on the GitHub repository and things like that, but I try to suggest different directions that people can take it in as well. So a lot of people bought it for specific applications. Like I mentioned, you know, the cavity detection and somebody else is doing the, um, prosthetic limb control. And, um, yeah, so basically there's like people that have specific things in mind for it. Um, but, um, but there's, you know, also somebody got it because they're really interested in inverse problems and they, um, and, and that's their, their, their, their passion is to, to try to get better, uh, results on these types of inverse problems. And this is a really classic inverse problem. Um, yeah, so, um, yeah, so it'll be super interesting to see what happens when people actually get ahold of it. Um, yeah, so there's about, how many, there's about six out in the wild right now, um, that sort of happened before the crowdfunding campaign. And so there's, you know, this, there's a new batch now that will, um, uh, be the, the, it was basically the crowd supply campaign was a kind of launch, launch the project.

Chris Gammell: That's great. That's great. Uh, where, where can people find out more information about the project? So first off, can they still back the project or is it kind of cut off now?

Gene: Yeah. Yeah. Um, uh, yeah, you can still back it. Um, just taking pre-orders now. So basically if you go to like a crowd supply and look up a spectra, you should be able to find it and you can get one. Um, and if you want to see, um, everything, there's like tutorials online at open EIT.github.io. Um, and so you can see all of that there. Um, uh, yeah. And so basically that tells you how to install it, uh, tells you about the sort of different modes and firmware that you can, you can try out, um, to, to get off the ground really quickly.

Chris Gammell: And are, uh, where can people follow you online if they're interested in learning more about the stuff you're working on?

Gene: Like, uh, you mean my Twitter?

Chris Gammell: Yep. That's usually what I'm usually what I'm asking and, and as, you know,

Gene: Oh yeah. I'm Jean tool, Jean, Jean tool on Twitter.

Chris Gammell: All right. Well, Jean, thank you so much for telling us about this stuff. I think that this is just, I mean, it's great that it's like a new type of technology out there and it's accessible and it's open. And I'm really glad that you're doing this stuff and you came to talk about it. Like it's, uh, it's a great, I think it's very electronics-y introduction into the world of medical. It's like, it feels like a really good merging of the two and, uh, I appreciate you're doing this stuff.

Gene: Thanks. Thanks. Yeah. I'm obviously really into this kind of melding of electronics and humans. So yeah, it'll be interesting to see where it goes.

Chris Gammell: Yeah. All right. Well, we'll talk to you soon.

Gene: Great. Okay. Thank you.

Speaker ?: Thank you.