 And now we come to the talk entitled Low-Cost Non-Invasive Biomedical Imaging. Current medical imaging has problems. It is expensive, it is large, rarely preventively used. And maybe you've heard of the story of a fMRI, this is the Magnet Resonance Tomograph. They put in a dead salmon and they can get a signal from brain activity from it. There's also lots of problems in the software as well. A little story, maybe you look it up. And how this whole mess can be solved with a technique called open electrical impedance tomography. This will tell us gene rental, a big round of applause for Gene. Today I'll be talking about an open source route for biomedical imaging using a technique that's in R&D called electrical impedance tomography. Not many people have heard of it, which is why it seems like it's important to mention. First of all, whoop, there we go. I'll just give you the vision of what it would be like if everybody had access to cheap biomedical imaging. Right now, you only get image when something's gone wrong. And moreover, you only actually get to use these tools when something's gone wrong in a first world country when you're lucky enough to be close to a hospital and have access to these technologies. That's a very limited number of people. What's even worse about it is it's hard to hack. So if you wanted to improve this technology yourself, medical physics is an amazing field. But it would be very hard to do so because you don't have a $3 million MRI scanner sitting in your garage. Maybe you do, but that's good for you, just not many of us do. If we did have cheap biomedical imaging, we could do things like do preventative scans. So you'd wake up in the morning, you'd take a shower, the device would be quietly imaging your body, would warn you when the slightest little thing went wrong, you'd do machine learning over it. It would be wonderful, wonderful for healthcare. So that's the vision of what biomedical imaging could be. And the other point is sometimes we move forward faster when we share the information. I worked in defense for a brief period and people didn't really share information between each other. And I think that inhibited science from moving forward. So sharing is caring. So today I'm going to go through a few different things. I'm going to go through the current biomedical imaging technologies. I'll give you an introduction to electrical impedance tomography. I'll go through the open source electrical impedance tomography project. Then I'll go through some applications that we could apply it to. And then I'll suggest a few different next steps that we can go into because by no means is it finished. Right now we have four different main existing imaging modalities. Your MRI scanner, which is a wonderful tool, it's huge, very expensive. The most commonly used imaging is actually CAT scanner, which sends x-rays through your body, which is ionizing radiation, which is bad for you because it causes cancer in the long run if you get too many of those scans. And it's actually the first scan that you'll get when you go into the emergency room. It's the most commonly used. And as we all know, we've got those grainy images that come from the ultrasound of fetuses. Wonderful tool except for the scattering due to the sound. Get scattered when you have different density materials next to each other. Not exactly an imaging modality, but a very important diagnostic technique is EEG. So you might ask, how do we classify these right now? We have three main types of resolution, spatial, contrast, and time. Spatial resolution is basically what space you can determine two different objects from each other. Contrast resolution is soft tissue or subtle differences in tissues. And time resolution, as it sounds, is how things change over time and how quickly you can do these images together. Your CAT scan, your basic machine in a hospital costs $1 to $2.5 million. You probably didn't get one for Christmas to play around with. Oh, well. It's also got this ionizing radiation. You've got a lot of maintenance and dedicated technicians. An MRI, say, your average three Tesla magnet with its own helium quenching chamber, no less, as well as dedicated technicians and experts who can actually read the images. Again, $3 million, an amazing and beautiful technology, but really expensive, amazing spatial resolution, the best. When it does something at this very high spatial resolution, it actually takes four minutes and 16 seconds, which is a really long time to take to do this wonderful spatial resolution image. Ultrasound, it's a bit grainy due to scattering. Average costs about $115,000, not too bad. It's pretty minimal health risks. EEG doesn't do any image reconstruction. In fact, it does very little in many ways, but it is still very useful. Average medical grade by EEG system is 40K. You might also know of some open source EEG projects, which are pretty cool. So just a note on the radiation of CAT scans. It's actually the biggest cause of radiation or contributing cause of radiation in the United States. So here I just put those biomedical imaging modalities onto a graph so that you can kind of think of them in terms of spatial resolution and time resolution, and where they fall in the picture of common things that go wrong with people, like X-rays or CAT scans are great for looking at bone and bone breaks, pulmonary edema, that's water on the lung, tuberculosis, huge in third world countries, massive problem. You don't actually need super high spatial resolution to be able to detect it. And it's important to sort of understand what you can do at different spatial and time resolutions. On the optimal goal of all of this, I put non-invasive electrophysiology. What that is is high spatial resolution and high time resolution. That's where you can measure ion activation or basically what cells are doing when they communicate with each other, which is right now only done in an invasive manner. Today I'm going to talk about this new technique called electrical impedance tomography and describe where it will fit in amongst what already exists. So what is it? Okay, yeah, basically you send AC currents through the body, say a 50 kilohertz current, and that will take different routes based on what tissue there is. So it might go around some cells and straight through others. And that's really important because differentiating, say, fat from muscle is one thing that you could do, but you can go further and differentiate, say, tumours from healthy tissue because tumours have different impedance spectra to the healthy tissue. So as you can see, that would be very useful to do. This setup here is called a phantom. What it is, it's like a simulated human body. You get some salt water, the body is 80% water as you might know. You get some meat or vegetables, you put it inside, and then you use that to image. So we have current flowing through all these different directions, and we recreate an image. Right now it's used for lung volume measurements. This is a baby with an EIT setup, muscle and fat mass. There's a paper on gestural recognition that just came out this year. You can look at bladder and stomach fullness. There's some research papers on breast and kidney cancer detection. There's another research paper on hemorrhage detection for stroke. You can also look at the, there's more R&D on the death of venous vision in surgery as well, which would be another interesting news for it. So all of these are sort of in the works. And you might ask, great, that sounds amazing. Why isn't everybody using it already? Well, yeah, it's really an R&D technique right now. And it has a big problem. Its spatial resolution seems pretty limited. So it's limited by the number of electrodes, but I'll discuss some potential ways to get around that as we go. It might not ever get to the spatial resolution of MRI, but maybe we don't need it to, to be useful because it's so compact, it's so cheap. Nothing about it is expensive. It's got better source localization than EEG. Yeah, it's non-ionizing, so it's not harmful to human tissue. It's also got great time resolution. So it has advantages and disadvantages. So I'll just remind you of what the first MRI scan looked like at this point in time. As you can see, it looks pretty crappy in 1977. And now it looks pretty awesome. That's a slice of my head, by the way, in a three Tesla MRI scanner. This is what early EIT looks like. That's with 16 electrodes only. And what will it look like in a few years' time? I don't know. I hope that MRI was, it gives you a pathway that EIT could take too. So now I'll introduce you to the Open EIT project. So the Open EIT project is obviously open source. It has a PCB design, done in Eagle CAD. It has firmware written in C, has a Python dashboard that lets you see the reconstruction in real-time. It also has a reconstruction algorithm, which I'll go into. And you can get it from GitHub right there. So how does it reconstruct an image? That's a, yeah. So Opening EIT right now has eight electrodes. And what you do is you send this 50 kilohertz current through every combination of those eight electrodes. And you get a different impedance value for each of those measurements. On the left, you can see basically what you're doing. You know where the electrodes are positioned. And so you get one value going horizontally. You add it to another value coming from another direction. And again, and you can sort of see it's getting a low-resolution image as it goes around adding those values together. If you use many, many views, you bring the image back. This is the radon transform, that's what it's called. And you basically just send lots of current through these slightly different angles. And you build up something called a sinogram, which is over there. And then you invert it to get the image back. I used OpenCV, which is a really common image processing library to do this. You can just do it with a regular image yourself and try it out. But what I did is exactly the same as what you'll do with a regular image, except I used current to be the input data. So this is the PCB design in Eagle. So basically it has a few different features. You know, a connector for your eight electrodes. It's running an ARM Cortex M3, which is quite nice. And it has a dedicated DFT engine for doing your Fourier transform in real time, which is also quite nice. A JTAG debugger to easily reprogram it. It's got a coin cell or external battery options. It has UART to get serial data off. And if you want, you can also flip it to Bluetooth mode and get the data off via Bluetooth if you felt like going wireless. At this point, you might be asking, oh, is this safe for me to play around with? Which is a really great question because the answer is actually, yeah, it is. There's some guidelines called the IEC 60601-1 Guidelines for safe use in humans. And basically, which says it should be, and open EIT is less than 10 macroamps, which is great because that's well within their guidelines. If you want to compare it to other things that are completely legal, say, I don't know if you've seen those late night TV ads for those ab stimulators that stimulate your muscles. There are about 15 to 20 milliamps just for reference and as a scale to look at the 10 microamps. So some of you might have used them already and that's, yeah, hugely more current than what we're putting through to image the body here. This is what the dashboard looks like. It does the reconstruction. You can connect to serial at baselines. You can obviously adjust sliders to look at the area that you want to look at. You can read from a file and fiddle around however you would like to. This is what it looks like when you reconstruct something. I have a phantom up there, which is a pot of water with a cup in it. And I moved the cup around anti-clockwise. So you can see in each of the pictures I move it around a little bit more and you can see the reconstruction there with me moving the cup around. Again, this might not be wowing you with the resolution at only eight electrodes. It's a proof of concept, but that's okay. Let's see if we can make this, I make this go, whoop, yeah. Okay, here's a real time video demonstration of it. Here's me with a shot glass. I'm moving around anti-clockwise. Hopefully you can see on the left the image being reconstructed in real time. And there we go. Move to the bottom. Wham, you can see it over there. And again, up to the top, you can see it over there. So that's a basic proof of principle version of it running. So yeah, the first MRI scan of human lungs wasn't that amazing. Early EIT scan wasn't either. Something else that you can use up for is differentiating objects, multi-frequency. This is what they're doing, the breast cancer and kidney cancer scans on. Basically you send different frequencies through those times called multi-frequency electrical impedance tomography. And you build up a spectrum. Here I've got an apple, a pear, oh no, a sweet potato, and some water. And I've sent through these different frequencies and I get these different spectrums. They're different. You can see that they're different. They're quite obviously different. But yeah, you can also just simply classify it. And on the left you can see where the water is, the apple is, the sweet potato is, or the sweet potato and the apple a little bit harder, that one. But that's basically what you do when you detect cancer. So that's what I did. But maybe we should look at the other papers and see what they did, because they did better than me. So there's this guy called Aristovich. In 2014 he published Spatial and Temporal Resolution using this technique, 200 micrometers less than two milliseconds, which covers most of the applications that I listed on that graph at the start of the talk. Downside here is that it was an intracranial array. So it was under the skull. So very dense electrodes, a lot more electrodes. I only used eight, he used like 256. So you can see that it can be, like the potential is there. So how should we use it first? What's a nice low hanging fruit? What about medical imaging in the developing world where I believe four billion people don't have any access to medical imaging? No MRI, no cascans. Why is EIT good for that? It's cheap to mass produce, super portable, super low power. So that would be a great place to start. What could we do first? So I'm gonna go back to this image again and have a look. So tuberculosis affects a lot of people in the developing world and you don't need amazing spatial resolution to detect it. That would be a good one. Or what about pulmonary edema? Pulmonary edema is water on the lung. It's actually already used for that. You can quite easily see the different volume present or the different conductivity maps it's called of a working lung and a not so working lung right there. Next steps. So what should we do to make this technique better? What should we do for open EIT to make it better? If you wanna innovate again, that's the GitHub project, just go ahead. Oh, that's an avocado, has a seed in the middle. Who knew? I do. Okay, so I see the two main routes forward as one would be this low cost biomedical imaging for the developing world. You could just stick with the static imaging reconstruction because why not? You'd need a few more electrodes than it currently has. One of the main problems with the technique is how you stick it to the skin. So my suggestion for that is why don't you just use a water bath and stick the body part of interest in a body of water because water gets rid of a lot of the, it's called a contact and pinch problem. Or on the kind of exciting science front, you've got the advancing neuroscience option which would be measuring both high spatial resolution and high time resolution. So that's the noninvasive electrophysiology solution or and that would be super awesome. There's a couple of ways forward to do that and I'm gonna sort of discuss each of those. So roughly there's physical configuration improvements that could be done. There's things that you can do to improve the spatial resolution. There's things you can do to improve the time resolution and then there's this interesting tack on at the end that I thought I'd mentioned which is write functionality. So we're using very small currents to read an image. What if we pumped the current up a little before you know it, you're writing. Think noninvasive deep brain stimulation in a focused way. That would be very, very cool. So contact impedance. Major problem right now. There is a well-known solution. I haven't done it yet. You do this thing called differential referencing, common mode rejection. Should be done, haven't done it. That's the next step. That means that it will work when you just attach it with electrodes on the body. What happens is electrodes have some capacitance and different amounts, which kind of interfere with the measurement that you wanna make, which you want to be very accurate in just of your body. You don't wanna include the electrode information in there that's changing. So there's a way to remove that. That's well known already. Another physical configuration improvement. Just increase the number of electrodes. Wonderful. Now you've just improved the resolution or placing the part in water. Another set of next steps would be on the mathematical side. I mentioned that I use linear back projection, which is a wonderful technique, which is that's how they do CAT scans with X-rays. That's exactly what they do. However, it makes some appalling assumptions like current moves and straight lines. That is not true. What you should do is get a finite element model and solve Maxwell's equations because current bends around objects. It's actually, it works in three dimensions too, which might not be all that surprising, but it needs to be solved for those three dimensions, which is why you just need to solve Maxwell's equations and create a finite element model. And there's quite a bit of work on mathematical solutions that get higher resolution. So that's another improvement area. And now I was going to mention this awesome new technique, which actually, there was a paper this year called Magneto, like, Acousto, Electrical Tomography. You might remember the FBI rule from high school. So when you have a current flowing perpendicular to that, there'll be a force. Now that force, say it's vibrating with 50 kilohertz, that's the AC signal that you're sending through. Now you have a vibrating compression wave. That's sound. You can pick that up with a little piezoelectric element, and that's actually a focus of work. From that, you can get really good edge information because, as I mentioned earlier, sounds scatters at edges. So you would also get the electrical impedance tomography information for the tissue sensitivity. Why not combine those results together and you would have a better tool? It currently gets lesser resolution in the middle, simply from how you, every combination of electrodes just ends up having a less dense number in the middle. You can also do something as simple as increasing the power that you send through if you're game to do that. This is a kind of gory picture. So right now, epileptics, if they're really troubled by their problem, which they are, often they go into hospital, have their brains opened up, and they stick this array on their head through their skull, and they leave it open for a week, and they try to induce seizures through sleep deprivation, and then they measure the activation potentials that way to locate the foci, or where they're going to do surgery to stop you from having seizures. But it would be much better and nicer if you could do it noninvasively, and you probably can if you improve the time resolution of EIT. There's nothing stopping you from doing that, by the way. You just have to, it's just a next step, really. And then I'll also mention right functionality. So there was a paper that came out halfway through this year by a guy called Nia Grossman, and what he did is he showed that you can stimulate neurons by sending current through the skull and in a focused way. Now, why that's interesting is you can noninvasively stimulate neurons, so that's the right functionality. It's unknown what resolution, or how well you could control the focal point here, but it works in the principle of beat frequencies. So he sent through two kilohertz and 2.05 kilohertz, and basically had a beat frequency of 10 hertz arise from that, and basically stimulated neurons in this area that he can control via an X and Y axis, which is very impressive, leaves a lot of questions open. So those are some possible next steps that it could go in. Obviously, I think this is interesting. I hope that you do too. I'd love it if you would wanna sign up to a mailing list. I'll give a link on the next page. If you wanna collaborate, email me. If you know any funding buddies that might be interested in the developing medical imaging for the third world, I'd love to be put in contact. If you wanted a kit, and if there were enough people that wanted a kit, probably of the next version, which would have 32 electrodes, sign up to the mailing list, talk to me. Thanks. Thank you. Thank you very much, and we have a little bit time for Q and A, and please, if you have to leave the room, make it in a very quiet way. So there are some questions. I've seen microphone four first. Please go ahead. So a great thing, thinking about developing countries and getting the medical tech. But at the very first beginning, you said, imagine a world where this imaging would be all available like every day, and it creeped me out a little bit. Do you really think that it is a good idea to go in the shower in the morning and have your, I don't know, your bathtub telling you that there is small maths inside your lungs? Thank you. So that's a good question. Basically, the question was there's a privacy concern with looking inside your body. Doesn't sound that great to some people. To those people, I would say you should turn off. I know that sounds a little harsh, but please just turn it off, don't use it. And with all scientific movements forward, comes great risk, I would also say, and it can be used for good or evil. And it's up to us as a society how we wanna choose to use it and how we structure ourselves and potentially motivate and incentivize corporations to use it in a responsible way. Part of making this open is a hope that basically if people have access to it, you can choose for yourself how you'd wanna use it. And next question would be from the signal angel, please. Yes, I have a couple of questions from the internet. First of all, what type of AC frequency is in use? Ask or assume sinusoidal, but he wonders if you also try square wave, triangular and other shapes. That's also a really interesting question. It's about what kinds of waves are used, what kinds of AC signals. Typically it's done with AC sine waves, ranging, they range all over the place for depending on what application you wanna use it for. I mentioned multi-frequency EIT for cancer detection. That uses a lot of different frequencies. So if you wanted to use other waveforms, I think that would be really interesting. If nobody's tried, you can, that should be done. Since there's a big queue on microphone three, I would go there, please. Yes, I have a technical question. Assuming that you won't use these techniques on humans or organic matter at all, what are the limitations for the resolution, the spatial resolution, and is there a possibility to reduce the spatial resolution? You mean increase the spatial resolution or reduce it? Reduce the, yeah, the voxel size. Oh yeah, yeah, so increase the spatial resolution. Yes, absolutely. So I was trying to go through a few of the next steps that could get to that. One of them is magneto-acousto electrical tomography because you get two different types of information which you could put together to form a higher resolution image. So that's one way. And if you didn't need to worry about human safety, I recommend you just turn the power up. That will also work. Thanks. Okay, I think we go back to the signal angel for one short, one please. Yes, I have another question from the internet from a doctor this time. He wonders if there are any clinical studies that compare the pulmonary edema diagnostics with EIT2 ultrasound and why don't we just work on cheap ultrasound instead? Yeah, that's a good question. So people are working on cheap ultrasounds. Ultrasounds gives different information to EIT and it has a problem of the sound scattering. So it's a different type of information which has different pros and cons and I think people should make cheap ultrasound and I would like to see the hybrid modality come together. You can get really good tissue distinction with EITs or there's pros and cons. Okay, then microphone two please. Hello, really good talk. My question, so far you always need direct contact to the electra, right? So it has to be direct contact or in water. Is there any way to detect or measure the signal without direct contact? So maybe if the object is in air or any other gas? Right, I wish there was, no, is the short answer. Any research on making it happen? Well, yeah, you can use X-rays. They work wonderfully to go through the air and but if you use them, I mean, you do increase your chance of cancer. So don't use them all the time on yourself. Again, CAT scanners are a little bit expensive. All right, thanks. Thank you and I think we have time for one more from microphone three please. Yeah, my question would be, maybe I've missed it, but what's the auto of magnitude for cost? So would this be feasible at like a hacker space for this to implement and does the industry see the possibility to make money with these technology? Yes, a lot of the sort of these early like R&D papers, yeah, they should be applied and you could make money with it, absolutely. And there's no component in there that costs more than a couple of cents. I suppose a Cortex M3 like costs a couple of dollars and I mean, I don't know what your budget is, but yes, I think you could do this in a hacker space without any problems. There's nothing stopping anyone from doing this and as we know microcontrollers are becoming cheaper and cheaper, so why not? So I don't get Hasty's signs from the sidelines, so I think I can take another question from two please. So far you have showed us images of 2D planes. Yeah. What about volumes? Yeah. Yes, so there's work on solving for volumes using finite element models and solving Maxwell's equations. Basically I just, and you can do that, basically I just did the shortest route to image reconstruction that was available which was linear back projection, which is typically done in a 2D plane. So absolutely you can do it in three dimensions. Yes. Thank you. So I'm very sorry, we are out of time. The queue back there, you can have the chance to chat with our speaker just right now. The next talk coming up is in about 15 minutes and it's I think also in English. See you then. And a big round of applause for our speaker. Excuse me.