Lecture 9: Laser Speckle Principles, Instrumentation, and Biomedical Application

Okay, so I guess we'll get started with today's, today's Speckle Day. Yay, Speckle Day.

So, it's my pleasure to introduce, uh, uh, the, the first speaker, Dr. Christian Cruz. So, Christian and I have a long history together. He was an undergraduate working, started working in my laboratory during his junior year, and then he, I couldn't scare him away then, so he stayed on for graduate school. He was actually one of the, he was, he was a participant of the short course in a previous incarnation of it that Bon and Jerry ran, is also an, a graduate research fellow. And some of the work that, some of the work that Yama showed in his presentation on the cardiac arrest CPR model with the, some of the, the optical imaging, so Christian, that was, that was Christian's PhD work. And so then after that, he stayed on as he was an NIH post-doc funded fellow, um, from the, uh, CSA that's here, and he has, is now a project scientist here. So, we've been working together for 12 years. Yeah. So, he, he's basically the lab illuminati. Everything that, that you see like in slides, so like that, he's at some, he's at, he's at some, he's at some, uh, influence on it, whether he's actively doing it himself, or whether he's advising, um, or working with the students in the collab. So, it's, and what he's going to be showing here, some of us have the work that he directly worked on, but this is like 5% of what he's done. So, um, I, with that, so it's, well, appr.

All right, thanks Bernard. Um, so hopefully, um, by the end of today, you guys will have a little bit more, uh, background on laser speckle imaging. I'll be giving the, the first talk, going over some basic principles, the instrumentation that's used, and then going into applications. So, uh, with that, as I mentioned, the principles, how we're able to utilize speckles to get blood flow information. And laser speckle imaging is a very versatile technique. We can use it in a lot of different embodiments, um, for, depending on the application that we're using it in. So that's the one key takeaway that, um, I'd like you to get from this first lecture, being able to utilize the principles of laser speckle imaging, but be able to apply it in many different scenarios. And the previous director at the Beckman Laser Institute, Bruce Tromberg, who's now the director at NIBI, one of the, uh, institutes at NIH, um, he stated that laser speckle imaging is one of the most simplest complicated technologies that we have, uh, at, uh, BI. And there's a lot that goes into this. And that, at its basics, LSI is very simple, and I'll be showing you what the, that technology is. But if you really get into the nitty-gritty of the physics that are associated with it, it actually becomes very, very complicated.

So, starting with the technology of LSI, it's, it's pretty simple. You need a coherent light source, uh, typically a laser is used, some sort of expansion optics to expand the beam onto your sample. This case, we're just representing it with a phantom, um, but it could really be any biological tissue. That light then interacts with the sample that you're illuminating. The backscattered light is then collected with a camera that's controlled with a laptop. So, not a lot of optical elements here, um, and it's pretty straightforward. And when we acquire, uh, this backscattered light with the camera, uh, it's effectively a random interference pattern, and this is what we call a speckle pattern. So, there are bright spots and dark spots in this image. And if you know what an old TV looks like when it's not working, it kind of looks like this. And we can utilize this information of the speckles to obtain blood flow information.

And I want to start with a little bit of trying to build intuition of how these speckles are formed. And to do that, one of the, the ways that I think is really building intuition is thinking of the light as it's traveling through the tissue and thinking about it as waves propagating inside the tissue. So, going to go through a little bit of a thought exercise where we have, uh, um, a packet of light with a specific wavelength. And again, we're using a, a laser, so it has a very, very specific, narrow wavelength, and we really are not having a broad light source. And we need that to be a very single frequency in order to have that interference pattern. And we're going to be looking at three different locations. This is representing a camera sensor. And one thing to, when looking at this wave is that we'll have an amplitude of one for this wave, and each pixel can have a maximum value of four. So, starting with the red pixel, if it's looking at four different waves, and you're looking at a single time point where this black vertical line is, you can measure the amplitude being at one for each wave. So, when you sum them all together, the value would be four. So, if our camera can only go between zero and four, this would be a bright spot on our camera sensor. So, since it's a bright spot, uh, based on just, uh, summation of waves, that's constructive interference. So, all the light is being added together.

So, now let's go to a different pixel on our, our camera sensor, the green one. And here you can see that packet of, uh, light number two and four are shifted compared to one and three, such that when you add their amplitudes up, it's summing to zero. And when you do that, that is, you'll get a dark spot on your image, and that's due to destructive interference. Okay. Now, it doesn't need to be all constructive or all destructive interference, but it could be somewhere in between as well, and that would appear as a gray spot. So, in this case, the intensity would be 0.4 if you looked at the single time point T, and that's really what speckle is doing. It's looking at these waves that are being collected by the camera sensor, where you have interference, whether it can be constructive, destructive, or somewhere in between. And why do you have these different amounts of constructive and destructive interference? Well, that occurs because when you send the light into the tissue, they don't all go over the same path. If they all had the exact same path, they would all come back to the camera and they all constructively interfere. But what happens when you send the light into the tissue, when it interacts with what's in there, they all go slightly different amounts of distances. So, that's where you get that variable distance. Some are still constructively interfering, some destructively interfere, and somewhere in between. And that's all because of how the light is going through the tissue. And from there, we get our speckle pattern.

Here's an example of using a coherent, hey, laser on two phantoms, uh, that is separated by this, uh, black vertical line, so on the right and left. And you can see the speckles. You see bright spots and dark spots, and you have your speckle pattern. Now, if you apply motion to one of the, uh, phantoms, you can see that there's this blurring that occurs. And what's happening with that is that this interference pattern is fluctuating very, very rapidly, and that is causing there to be a blurring phenomena to occur. Yeah. Uh, so this one, you, uh, so, uh, the question was, what type of motion do we get to the phantom? So, if we had these two phantoms, all we did was move it forward. So, it was just moving it, and that was causing there blurring to occur, uh, in the phantom left, um, and that's due to the interference pattern fluctuating very rapidly.

So, just to try and build an intuitive example, in this, uh, image, we have two objects. One is a train that's moving, and the other is a person. So, if you're taking a picture of this, right, if something's moving very fast and your exposure time or shutter speed is long, uh, then it's going to have a blurring effect, where the person that isn't moving, they're very sharp in their features, and there is no blurring. So, you can intuitively think about, uh, when you're taking a picture of these speckles, when the, uh, you have something moving, it's going to be a blurring effect.

So, now, um, with that, some basic principles, going to go into how we simulate flow in, uh, a lab, laboratory setting. Ezekiel has run a lot of these in the lab, um, and, uh, what we have here is we have a phantom that there's a tube embedded into the phantom. And we would have a syringe pump, uh, that has some sort of scattering fluid in it that's being flown through, through the tube that is embedded in the phantom. We're illuminating, uh, the tube and collecting the backscattered light with the camera. So, once we do this, we can acquire that raw speckle image, and we have that tube with scattering fluid going through it. And you can see that there's blurring within this image, where areas that are static, you have that high contrast, which we'll get into in a moment, but basically, you see lots of dark and bright pixels compared to the middle, which has that blurring effect.

So, now, how can we actually convert this blurring to information that, uh, can be more quantitative? And to do this, we use a simple sliding window technique, as one of the common approaches. So, to do this, um, the, uh, window here, that's 5x5, is our local window where we can, uh, compute our speckle contrast. And that's this KS term here, standing for spatial speckle contrast. And within that window, we compute the standard deviation of the intensity values that are collected by the camera and divide that by the mean intensity. So, here we have an example of intensity values, uh, that are within this 5x5 area on our camera sensor. And taking all these values, you can compute the standard deviation, divide that by the mean intensity, and get this value that's 0.049, and that's what we call our contrast. So, that would be done in one area of the image. But how do we get it over the entire image? Well, we start moving this sliding window and we repeat the calculation as we go through it. So, here's just an animation where we have 81 pixels on a raw speckle image, and we are going to have our 5x5 window. And then we're not replacing the value in our raw speckle image, just to make that clear, but it's being assigned to basically a new matrix of speckle contrast values. And then when we slide this window throughout our whole image, we're building up our speckle contrast image, uh, defined by that standard deviation divided by the mean intensity. Yes.

Yeah, so there are different ways that you can account for that, depending on your sliding window algorithm. Sometimes they fold the image over, but in general, we have so many pixels in our image, you know, we're like 1000 by 1000, that removing like a two-pixel area around the border really isn't going to do a lot to our images. But there are ways and algorithms to kind of account for those edge effects, depending on what you're doing. Yes. Yes. Yeah. So, um, the question was, can you alter the, the sliding window size? Um, so yes, you can alter the sliding window size. Um, okay, so we'll say that for the lab later.

So, that's how we're able to calculate the speckle contrast. And, um, now going back to this phantom exam example, and applying that sliding window technique to the image, you can see that in the middle, it's very dark. We, on the, uh, where it's static, it's much brighter. So, this speckle contrast term, the lower the value, there's more motion in, uh, that area. And when the speckle contrast is higher, there's less motion. Um, and one thing that you can think about, if you're looking at the intensity values within this tube, I remember the contrast is the standard deviation divided by the mean, is that all these pixel values that are blurred have very similar intensity values. And because of that, the standard deviation goes down. So, that's another intuitive way that you can think about why the contrast becomes lower when there's more motion, because those intensity values are more similar.

Now, that was an in vitro example, but we want to use this technology in biological tissue. And what is that scatter that's in biological tissue? That's red blood cells. So, obviously, all of us have, uh, blood flowing through us. And these red blood cells can scatter that light, uh, to induce, uh, intensity fluctuations in the speckle pattern. So, here's an example where we, uh, illuminated the, uh, rat brain and collected the backscattered light with the camera. And this image, you really don't see a lot of features. You can maybe see some blurring in some of these areas, but for the most part, it just looks like noise. Now, when we apply that sliding window algorithm, you can really start seeing the accentuation of blood vessels that are occurring in these areas. So, um, all these vessels are from the cortex of a rat brain. So, it's really showing that you can have this high visualization of the vasculature by doing, uh, the sliding window approach.

And then what we normally do is that we convert the speckle contrast to a term called speckle flow index, uh, which is inversely related to the flow, to the correlation time, and it's proportional to relative flow speed. So, in this image, areas that are red have higher flow, and areas that are blue are lower flow. And just to give a couple of examples, um, that it was used in rodent brains. This was actually one of the first main applications and demonstrations of speckle in the brain, where you can see that there's this increase in blood flow that starts in this area and moves down to the left, which is associated with seizure-like activity called cortical spreading depression and its propagation. Another one that we use in the lab quite frequently is changing what the, uh, rodent is breathing. So, in this scenario, it's a 15-minute experiment. The first five minutes, they're breathing room air. Then their, what they breathe switches to 5% CO2 with the balance of room air, and then switching back to room air. And this is a, a time-lapse video of that occurring. And you can see right around the 5-minute mark, there's this large increase in the blood flow, uh, in this image. And then we can quantify it by taking different regions of interest in the image. And right at that five-minute mark, you see this large increase in the flow. And then when we go back to room air, it starts decreasing slowly. So, those are just two brief examples of, uh, laser speckle imaging. And from there, I'm going to take a little break to see if there's any more questions.

Yes, good question. Um, so the exposure time is very important. Uh, so the question is, what was the exposure time for this data set? Um, so for this, we use an exposure time of 10 milliseconds. And I don't remember if there's any lab associated with exposure time, um, but, um, I'll just briefly talk about, well, there's multi-exposure, right? Or no, in lecture. Okay, so Bernard, we'll get into it a little bit later then about the exposure time. But for this, we use the 10-millisecond exposure time, which is important for tuning the, the sensitivity to flow. Yes.

Uh, so the question is, when the intensity values are more similar, is the contrast, uh, uh, lower or higher? Um, so when the intensities are more similar, uh, then the contrast goes down because one of the terms in the contrast calculation is standard deviation. So, if you have much more similar values, then your standard deviation goes down, which then decreases your contrast. Yes, that's a good question.

Question. So, inherently, a laser is more polarized. Oh, sorry. Um, so the, the question was, um, do we use the polarized light when sending it into the, the tissue? Um, so, uh, inherently, the laser does have pretty good polarization, but we typically do, uh, polarize the light, um, with a polarizer. And then on the detection end, we actually have a cross, uh, cross-polarized detection to minimize surface reflections. Um, so basically, specular reflection, uh, can cause issues, because it will just be saturated, and it will artificially have a low contrast, because if your detector is saturated, all the values are, say, in an 8-bit detector, 255. So, you don't have any resolution to resolve anything. So, yes, you want to polarize light going in, and typically cross-polarized light coming out to minimize any specular reflection. Yeah. Yeah.

So, the, typically, uh, so the question is, uh, is the, the laser need to be coherent? Um, so, as narrow of a line width as possible for the laser is ideal. Um, there are practical limitations to that, because the more coherent your laser, the more expensive it can get. So, there's been a lot of papers that obviously used a very expensive, very long coherence laser, like the, which has hundreds of meters in terms of its coherence length. So, just using a very inexpensive laser diode, which, you know, is still coherent, but it's much, much shorter in terms of its, uh, coherence length. So, it does depend on your application, and what you would want in selecting your laser. I'm not everyone can afford, you know, a $10,000 laser that's really long coherence laser, but do need that. And my answer would typically be no, you don't need a laser that expensive. Yes.

Yeah, sure. So, uh, in this, uh, it's basically taking from, uh, this image where we have our contrast. So, we generally will have a, a large, what we call, kind of a global ROI, um, where we're taking areas where we see vessels, where we see no vessels, and it's more of like a total perfusion measurement. Um, and what we do is at each time point, we have a raw image, we convert it to a speckle contrast image, and then with everything in that region of interest, we take the average value out. And then that would be like one of these time points on this graph on the left. And then, so that would be the large ROI. And then we have the small ROI, which is just focused on the parenchyma, so it's much more capillary-based. And the brain is actually very vascularized, so it's really focusing on the capillaries, um, and doing the same sort of thing. So, all the values within that small ROI, extracting out the speckle, or the, the flow information, and then plotting that as a function of time. Yeah.

Yeah, so the, what the question is, what is the, the color scale here? So, this is that speckle flow index. So, it's inversely related, uh, to the correlation time, um, and it's more indicative of relative flow, um, that is occurring in the, the sample. So, speckle flow index. Um, so the, the question is, how do you calculate speckle flow index? Um, so, just briefly, it's SFI, where speckle flow index is equal to one over two times the exposure time of the camera times the contrast squared. So, you can do that at each pixel location. And the reason for that is a simplified approach to calculate, uh, for calculating correlation time, where that term, two times exposure time times contrast squared, is basically correlation time, and speed is one over the correlation time. So, that's where that, that term kind of comes into. Yeah.

Uh, so with that, it would just be, uh, we normalized. The question was, how does true blood flow percentage is calculated? So, basically, we have our, uh, SFI flow index map. And then over the entire baseline period, so that first five minutes, that would be our baseline SFI. And we normalize everything to that period. Um, so each SFI value normalized to, uh, the baseline SFI, and then multiply by 100 to get our relative true blood flow value. Does that answer your question? Okay. Yes.

So, sorry, can you repeat that? So, the question is, do you utilize basically all the pixels, or are you selectively choosing, like, the area to get your baseline flow value? So, for a lot of our studies, you can normalize basically pixel by pixel, across time. So, you kind of get, like, a relative map of flow. For a lot of the data that we do, we will kind of take the flow within the ROI and then, um, use that information. So, basically, we, we extract the flow information, we plot that as a function of time. And then from the time course data, so we're not really, we extract the, a single value, and that's what we use for the baseline over, say, this experiment, experiment from zero to five minutes. Does that help? Maybe we can talk a little bit more about, uh, some of the nitty-gritty. Okay, a lot of good questions.

Um, oh, there we go. Um, so, um, the, the first, uh, application. So, now I'm going to get into the applications that, um, have been done in the lab. And really, the, the first application, I believe, that Bernard started in the lab using laser speckle imaging was to look at port wine birthmarks, and especially trying to monitor blood flow during, uh, treatment. And one of the, the big, uh, people in the field of port wine birthmarks for treating them is Stuart Nelson. He's the director of the medical clinic at the Beckman Laser Institute, and he really was, uh, super critical in developing the device that allows there to be, uh, effective treatment for port wine birthmarks, which is a pulsed dye laser. And, oh, okay, I'll go, what port wine stain birthmarks are first. Um, so it's a, re-engorgement of the tissue, that progressively worsens over time with age. And what this re-engorgement is, is a hyper-dilated blood vessel network. So, you have a lot more vessels than really should be, uh, there in the tissue. And as, uh, if it's not treated, you can can start developing these nodules, on the skin, which actually are extremely painful.

Now, when you can actually treat it using, uh, a pulsed dye laser, which Stuart Nelson was instrumental in creating, you can start seeing that the port wine birthmark is starting to lighten its tissue, uh, its color. And you can also do additional treatments to, uh, laser treatments to, uh, decrease the nodules, that are, uh, there. So, so what is the laser treatment really doing? So, it's using a pulsed dye laser, pulsed dye laser, to selectively heat the blood vessels, that are in the skin, and with the goal of trying to completely coagulate the blood vessels. And what is the reason for, uh, this coagulation? When you have, before treatment, you can see the redness. And then going into subsequent, many, many treatments of looking at the, using the pulsed dye laser, you still have this, uh, re-engorgement of the skin. So, why is this not occurring, the complete lightening of the skin occurring, even after so many treatments? And that's, um, because if you have complete shutdown of the vessel, you can have a good clinical outcome, which is that lightening of the skin. So, you completely shut down, you have limited angiogenesis, and you have good skin lightening. So, what happens when you don't completely coagulate the blood vessel, or it's not a good clinical outcome? Well, the vasculature can still be very, uh, dense. You can have additional, uh, angiogenesis, and you do not have that skin lightening, uh, with the port wine birthmark. So, effectively, there is not a good way to monitor the treatment to try and make sure that there is a complete coagulation of the blood vessels to try and have good clinical outcome.

So, with what, uh, Bernard, uh, did to try and look at the efficacy of treatment, was putting a laser speckle imaging, uh, device, and bringing it to the clinic during the port wine, uh, birthmark treatment. So, this is a cart-based system where we'd have a computer that has real-time processing. So, it would acquire your raw speckle image and convert it to the speckle flow index map, so you can see it in real time. Here is an example of a patient, um, undergoing treatment by Dr. Nelson, and having the laser speckle imaging, uh, image the, the subject. So, this is an example of the LabVIEW GUI that would enable, uh, real-time processing of of the speckle images to look at how the laser therapy is going.

Yes. So, I'm not an expert on this, but I'll try my best. So, the, the question is, how much, uh, time or how many passes would be to knowing that they're done with the treatment, basically, right? So, when they would, um, some, I believe, when they would do the treatment, the doctor could look at the amount of, I believe it's called purpura, that would occur, when doing the laser therapy, and it's really a visual assessment. Does this look like vessels are kind of the skin turn more purple, which would correspond to shutting down the blood vessels? So, there really wasn't a functional measurement, and that's where, uh, experience from the treating physician is really important. Dr. Nelson has really good at this because he has so much experience with it. But you couldn't take average Joe and say, here's a device and start treating them, because they don't have that clinical knowledge of when the therapy would be sufficient. So, the goal with trying to add laser speckle imaging is to try and have more of a clinical, uh, and functional measurement of that shutdown of blood vessels.

Um, so this is an example where you have increased perfusion in the area that the port wine stain is in. They can do multiple treatment passes with the pulsed dye laser. And after the first pass, you can see that there is a reduction in blood flow. Now, one thing I would like to mention is that the clinician did not, was not able to look at the laser speckle imaging during treatment. They did complete normal standard of care. So, these measurements were taken more of, let's see what the clinicians do, and how does that correlate with the, the outcome.

Yes. So, typically, the LSI, most of the entire signal is within the first 500 microns to 1 millimeter of of tissue. It'd be really rare. So, it would be much more at the capillary level is what's being resolved. So, especially in skin, it's more of a perfusion measurement, where some of the rodent brain, you can actually see some of the larger vessels because they're actually exposed. What's that? Oh, uh, in the skin, um, during the treatment. So, you're asking about if there's any clots during the treatment? Uh, I am not 100% sure on that. The skin is fairly, um, at least compared to the brain, is really low vascularization. Yeah, I would think so. Um, but again, a lot of the targeting here is that the, the smaller blood vessels that were able to monitor with LSI, than like the deeper arterioles, especially the, the skin. It's a good question.

Um, usually it would, it may offer some benefit, but due to the scattering of tissue, um, it's going to be really challenging to do that, um, at least with the imaging-based method. There are point source techniques, diffuse correlation spectroscopy, if anyone knows about them, that are able to resolve flow much deeper, uh, than laser speckle imaging, which is a much more superficial technique than those point source measurements. But here you have an image that you can deal with versus a single point source measurement. So, the question was, yeah, yeah, exactly what Von said.

Um, so, uh, after the first pass with the pulsed dye laser, there was a small reduction in flow, and the clinician thought this area needs to be treated again. And with, through the second treatment, there was a much larger reduction in terms of the blood flow that occurred that was visualized with LSI. And then one of the things that we wanted to to see was, is does the decrease in blood flow correlate with the skin lightening? So, basically, the port wine birthmark becoming lighter after treatment. Um, and the, the result was that for the large majority of subjects, the decrease in blood flow, so more of a decrease in blood flow, correlated with more of a lightening in the skin. So, again, this was done without the clinician knowing about the blood flow values, and there was a decrease in blood flow that was correlative with a lightening of the skin. These four, uh, subjects here were, um, even though they did have a decrease in blood flow, their blanching, or their skin lightening, really didn't change much. The exact reasoning was unclear, but going back to kind of the different pathways that can occur due to the coagulation, maybe these subjects had more angiogenesis that occurred following treatment, than the ones that did respond, uh, quite well.

So, going into another, uh, clinical-based application, where having the laser speckle imaging system on a, uh, cart-based system, where it's on an arm, where you have the bler, um, that is on the arm, and then that light can then be collected, um, when it's interacted with the tissue. And this application was from a group out of Vanderbilt, where they were looking at parathyroid, uh, looking at the parathyroid during laser speckle imaging. So, one of the things that is important with looking at parathyroid viability is that they need to know whether enough blood flow is getting to the parathyroid to know if it's doing its function. So, in this top panel, A through D, the contrast here, you can see that it is relatively low, in comparison to when the, uh, the parathyroid here in the white circles, the contrast is much higher. So, just to reiterate, lower contrast, more blood flow, higher contrast, less blood flow. So, this was a, kind of a first approach to be able to show that you can use LSI to look at blood flow within the parathyroid, or if there's not enough blood flow going to the parathyroid.

And going into, kind of, the first two systems I showed are very cart-based systems, going into the clinic. And there are commercial LSI devices. We have used this Pericam, uh, LSI system, which, you know, it works pretty well from everything we can tell. If you have, um, quite a bit of money to spend on it. I believe it's, last I checked, about $200,000. Might be less now than a few years ago. But one thing that you can see with these clinical devices is they're all on a cart. They don't really aren't conducive to be able to be super mobile, um, if you're in a scenario such as the NICU or in the army.

So, before going into that, I'm going to pause here for any more questions. Yes. We'll see. Lymph nodes. We have not. Um, so the question is, have we ever looked at lymph nodes using LSI? Um, we have not. My one concern with the lymph nodes is, based on my understanding, the flow is much slower than traditional blood flow. Um, so maybe if it's exposed, it'd be able to be resolvable. But there are obviously practical limitations with that. But it might work. Um, it also, I'm not 100% sure about the scattering element, uh, that's in lymph. If it has, you know, similar to to blood, it's able to scatter the light so we can see changes in the interference pattern, if the, the lymph also has properties that are conducive to that. Yes.

Uh, this one here, I believe, wait, this one, this one here was actually the, the laser to do the speckle. This laser was more for positioning, I believe. So, knowing kind of where you're shining your laser, it allows you to see, hey, I'm in the right position, because you don't want to be too far out of focus, one way or another. That knowing where that laser pointer goes, it's able to say, hey, we're in focus, we can collect data at this location. Well, the, so the, the question is, how many images do you typically use for the temporal imaging? And the answer, it varies a lot. Um, so sometimes, you know, you can get away with, say, 10 images if you want to look at just a small area in time. Sometimes, you know, you want to use a second worth of data to give a really high-resolution image. Some of that you will be covering in the lab later today. Um, and then there's other times when you just want to use a single image, and we'll get into that a little bit later, of what you can utilize with just a single image and looking at that very quickly over time. Okay.

Um, so going into, kind of, as I, I briefly mentioned, with current LSI devices, they're very bulky, they're expensive. And if a lot of our work is funded by DOD, as well, where they really don't have a lot of space for a large clinical-based device that can be wheeled around super easily. Um, we also have done some work in, uh, the neonatal intensive care unit, where, you know, they already have a ton of monitoring devices. It's a very special population where you don't want to interrupt any of their workflow, especially, um, if the, the baby needs help pretty much right away, you don't want to have to move your device, uh, very quick. So, you want it to be mobile, but also be small enough that you can target these areas. So, um, one thing that, uh, Bernard and, uh, previous grad students started doing was kind of miniaturizing that huge cart-based system and be able to have it in just a handheld format, where you're able to attach, uh, a laser and the camera, and lens to a surface. And this was basically the whole device. So, you can just hold it, it'll be able to acquire the data for you, and have a, a decent size field of view, um, in this handheld format. And one of the things that I mentioned going into the NICU was that there, there was a neonate that had compromised blood flow to their thumb. So, that's shown in this area, and laser speckle imaging was able to detect it. This was work that was done with Mustafa Kabir at the Children's Hospital of Orange County, in Orange, um, so fairly close to us, but able to detect these flow-based differences in this kind of, uh, niche population, but also is very important.

Um, kind of going into some of the validation work using these handheld-based systems, was using a burn wound model. Actually, my first projects in the lab were using burn wound-based models to look at blood flow. And there's different classifications of burn wounds. Kind of the two extremes are superficial, which is kind of like a sunburn. If you ever have had a sunburn, it's pretty uncomfortable, but typically you would have an increase in blood flow when you have that sunburn. And full thickness is basically going completely through, uh, the tissue and starting to get into the lower dermal layers. And when you have a burn that's that severe, you generally will have a decrease in blood flow. So, using the handheld device, it was a proof of concept, being able to detect what you would see with a more traditional, uh, mounted laser-based, speckle imaging system, where you can see that there's an increase in perfusion with less severe burns, and a larger reduction in flow when you have a more severe burn.

So, now we have this handheld device, um, that works pretty well. But what can we do to kind of go into the next generation? And some of the different modifications that were made was to remove the tablet from the system from being on the handheld part. The field of view was, uh, larger. In addition, instead of having these, uh, Thorlabs, opto-mechanical components, which do add up in terms of weight, we opted for 3D printing, different components that are much more lightweight. And then the last part, which was the, the key innovation, was that the, uh, the laser and camera was placed on a stabilizer. And why would we want to use a stabilizer? Well, the, uh, stabilizer is actually commonly used, uh, when filming, to reduce motion artifact when someone is holding, uh, a camera. And you can also easily change what you're looking at. So, here's a video where, uh, Ben, a previous PhD student, was able to control what the LSI was looking at in just a handheld form, using the, the, the stabilizer. And to compare the previous handheld to the stabilized version, a lot of work that Bernard did, uh, in the lab was using this dorsal skinfold window chamber model, where the epidermal side is removed on one side to expose the vasculature on the back of a mouse. So, we wanted to use this to compare, kind of, the vasculature in a handheld format and comparing it to the stabilized format. So, on the left, we have the handheld images, and on the right is the stabilized. And one of the, the key things in going through this video is that you do see a lot of motion still, but for the most part, the stabilizer, you're still able to resolve the, uh, vasculature from the background, as you're going through time, where in the handheld, very, very few of the frames actually are able to kind of get that clean vasculature compared to the background.

Yes, that's a, the question was, what was kind of the working distance? I don't remember what it is off the top of my head for this system, but with, with any LSI system, there is a variable working distance depending on what you're using it for. You can be, uh, couple feet away, or be like almost a microscopic, uh, view distance away. Yes.

So, for this system, it was just to collect the data, um, but we did have the tablet basically right next to the system. It just wasn't being held onto the device. So, it just made it lighter overall, because it was just holding the stabilizer now, instead of holding the tablet and the laser and camera and all the other components as well. Okay.

Uh, I know there were a couple questions, but are any more questions on, kind of, the changes in form factor going from the, uh, clinical cart-based system down to these handheld formats? Okay.

So, now, which, uh, some of the work that I'm definitely a lot more interested in lately is starting to look at the pulsatile component of the waveform and how we're actually able to get that. So, before, and I, I briefly mentioned it earlier, is that, you know, we, we can get this, this entire image and be able to look at spatial information with laser speckle imaging. But now, if we were to just look in one area, and we're using effectively every single frame and converting that to a speckle contrast image, we all have our hearts going, all of us, right? So, what we can actually do is start resolving the pulsatile component, which is related to each heart contraction. So, here we're able to resolve the blood flow changing with each contraction of the heart. And what we call that is a mouthful, but speckle plethysmography, or SPG for short. And, um, one of the things in that looking at the SPG is that if you've ever seen a pulse oximeter, it outputs a very, very similar waveform. And that signal is called the photoplethysmography, or PPG.

So, going into, kind of, an image of looking at the pulsatile component over time from a wrist, you can see that with each, uh, basically contraction of the heart, there's this large increase in blood flow that is able to be resolved, in a non-contact, widefield, uh, approach. So, each of these pulses is related to the heart contracting, and we're able to see that in many different areas over the body. So, those two, uh, examples were in a non-contact-based format. Here, I'm now going to go even smaller in terms of the LSI devices. And now I mentioned pulse, pulse oximetry, you know, everyone probably knows that pulse oximeter clips onto your finger. Well, we wanted to kind of drive towards having a finger clip-based device. And this, uh, device was actually created by previous grad students and postdocs that were in the lab. They had a spin-off company, with this device, that was FDA cleared, called the FloMet. And within this finger clip, I mentioned earlier, there were questions about, uh, kind of the coherence length of the, the light source. Um, and now we're just using a very small laser diode that's embedded into a finger clip, and another, uh, small component with a camera that's embedded into the same clip. So, now, and you'll actually get to see, uh, this device later, in the afternoon session, where you can actually clip this onto the finger, and you have that light source and the camera, all within this really, really small form factor, to look at, uh, the blood flow, and especially the pulsatile component of it.

So, kind of, what does the, the signal look like in this finger clip-based format? On the top, we had the SPG signal, and on the bottom, we have the PPG signal, which is the, the, the main signal, uh, in, uh, pulse oximeters. And you can already see that there's differences in their waveforms, but that makes sense because they're inherently measuring two different things. The SPG is more looking at the flow signature due to the speckle fluctuations, where the PPG signal is really an absorption-based contrast method. So, um, you can see that the SPG is a much more sharp peak. This down here is the dichroic notch, where the PPG is is much more broad over time. It still does have a dichroic notch, um, but much, uh, smaller, uh, resolution before it decreases.

So, one thing that we wanted to do was to compare the SPG waveform to the PPG waveform in terms of signal quality. So, in this graph, we have black is the SPG signal, so from laser speckle imaging. The, uh, dotted blue line is the PPG signal, so based on the main signal for pulse oximetry. And if you look at these, if you're just to look at the curve, you say, hey, they look pretty similar. But if you actually narrow down and look at the axes for the PPG compared to the SPG, there's actually a 40 times signal increase for the SPG signal compared to the PPG signal. So, much, much more robust signal. And we've been able to, kind of, test this SPG waveform in a few different scenarios. So, one of the scenarios was a cold pressure test. So, basically, you're, the other device on one hand, you put your other hand, uh, in a bath of ice water, and it typically causes phase of constriction to occur. So, if everything is normal, then you see that there is good signal quality in the SPG, as well as the PPG. But as soon as you start having vaso constriction, the SPG signal, which is in blue, is still quite robust. You're still able to completely resolve the heart rate. However, the PPG signal becomes quite noisy over time, and it's really, really hard to, uh, utilize it for anything unless you actually do a lot, a lot of signal processing, where the SPG is very, very little processing you need to do, and sometimes you don't even need to do anything at all.

Yes. We have not put two devices separated apart to estimate blood pressure. There is work being done by us and others to use the SPG signal, basically, to look at blood pressure. Um, we haven't done it to look like at a pulse transit time, for example, which may give additional information. But I think that's something that can be done, definitely. Um, I think it's worth investigating because obviously non-invasive blood pressure is a quite large field, that, you know, potentially having multiple LSI-based devices to look at this could do that. Yeah. Um, it's definitely something we're interested in. It's been a bit more challenging than expected, um, with some of the factors that go into it, but I definitely think that there's still an avenue where SPG can be used to try and get non-invasive blood pressure.

Another test that we wanted to do, uh, was, um, uh, was, uh, motion. So, can we actually have the SPG device, um, and have the subject basically just move their arm up and down like this, and compare that to PPG? Because one of the main issues with PPG is that it does, it's impacted by vaso constriction, but it's also impacted by, uh, motion. So, at rest, everything looks good in both the SPG and PPG, as expected. However, once you start introducing just vertical motion in this, your hand, the SPG is impacted a little bit by the motion, as you can see it kind of going up and down. However, the PPG becomes really, really hard to see the waveform features that you'd want to see, uh, in order to resolve anything like arterial oxygen saturation. So, using the finger clip-based device, it does appear that it is more robust to.

motion artifact. And then, going back to temperature, this was just in the cold room. Um, sometimes in the hospital, the patients can get quite cold. Um, and this was just a room that was at 74 degrees. We had a room that we lowered the temperature to 67 degrees and we performed the same measurement. And in blue, the SPG is still quite robust even with this decrease in temperature. And it's not like the room is that cold either, it's just, you know, a drop of, uh, eight degrees or so, seven degrees. But the PPG signal does become quite, uh, noisy over time to where it's really, really hard to determine what is the actual peak that you should be using in any of your calculations.

And one of the things where we think that the SPG could have a really big, um, impact is, um, looking at racial bias. So, um, with individuals who have increased melanin in their epidermal layer, pulse oximetry is known to inaccurately measure their oxygen saturation. So this was a, a big publication in the New England Journal of Medicine where on the x-axis you have the oxygen saturation that would be from a pulse oximeter. And then on the y-axis, you'd have the arterial oxygen saturation, uh, from, uh, blood. And just focusing, for example, on this group that's, uh, at, say, 90% oxygen saturation for the pulse oximeter, um, the arterial oxygen saturation is actually much lower. It's below kind of their threshold of, of 88, to where someone would require intervention. Um, so, but if you're below, above the 90, typically they, you know, they need a monitor to the patient, but no direct intervention would be needed immediately. And effectively, the pulse oximeter is saying someone has a higher oxygen saturation when they really don't.

So one thing that we wanted to do was look at the PPG signal in comparison to the SPG signal in different melanin concentrations. So this was modeling work, uh, by a previous, uh, uh, PhD student. This work, I believe, was done with Carol, um, in this with Caitlyn, um, where they were looking at the pulse oximeter, uh, the PPG signal. And if you have a melanin concentration in the tissue, uh, the signal reduction with the PPG signal was about 45%. However, when looking at the SPG signal, the LSI, um, arm, there would be a reduction of less than 10%. So very, very large distance, uh, difference between what it does to the PPG signal compared to what it does with SPG. So this was modeling work, and this was actually shown in vivo, uh, by the founders of the, uh, that finger clip-based device, um, and they had an indirect observation that this is occurring and that the SPG is much more resilient to, uh, the effects of skin pigmentation.

So one of the things that we wanted to look at was, so we know that, um, the SPG does appear to be more resilient, but do we need to account for the optical property changes with the SPG signal? And to do this, we need to modify the finger clip-based device, uh, to enable the correction of that SPG signal when there's changes in optical properties. Uh, so that being the absorption and scattering of the tissue. Um, and the second thing that we needed to do was have an experimental design where we can dynamically change the optical properties within the same subject, so we can see, uh, how if we account for it or don't account for optical properties, how the SPG signal changes. So to do this, we, uh, made a low-cost device where this would be a housing. We called the technology coherent spatial imaging because we're using the coherent light source and we're looking at different distances, um, from where the light source is and where the light is being collected based on the camera, um, so CSI basically combines two techniques. I'm not going to get into, uh, SRDS much, but it's basically resolved diffuse reflectance spectroscopy, um, which is a method that you can use to extract optical properties. Um, and then there's LSI, which in the light source, we're using a coherent light source, and this is able to measure the optical properties and being able to correct the blood flow from the optical properties as well as uncorrected. Um, and I believe this will be covered more in the, the next lecture. Um, this is a depiction, or this is a real-life photo of the CSI device that is on the finger clip where all this housing is, uh, being shown.

So in order to change the optical properties, one of the things that we were looking at was doing changes in blood volume. Um, so if we were doing a venous occlusion, um, we, or in this venous occlusion protocol, we would have a two-minute period of baseline. And then for each subsequent minute, we would be increasing the, uh, occlusion pressure by 20 millimeters of mercury and holding it there until we get all the way to 100 millimeters of mercury for the last minute and decrease it. And effectively, what's happening is with this venous occlusion is that you have blood coming in because our occlusion pressure is lower than the arterial pressure, so that would be typically 120 millimeters of mercury. But the occlusion pressure, especially once you start getting to 60, 80, and 100, it is lower than the venous occlusion. So effectively, we're blocking return back to the heart to be, um, go back to the body. So you're getting a, a pooling, uh, effect, basically, of increased blood volume in this area.

So what do we see, um, uh, once we do this? So there's, uh, the red curve is where we're not doing any correction for the optical properties, and we have the different occlusion pressures. And then the blue is once we actually correct the, uh, the blood flow for the optical property changes. And you see a couple of different, uh, interesting features. Starting around 40 millimeters of mercury in terms of their flow value, the, uh, uncorrected actually still seems like it's elevated, where once you correct for it, you can already start seeing decreases in the blood flow due to blood volume. So that, um, intuitively made sense to us. It gets more apparent as you increase pressures, um, where you start seeing larger separations. And the second thing that we observed was that the SPG amplitude, so even though this looks like noise, when you do zoom in, you can see the actual pulsatile waveform, that SPG signal, um, that the amplitude is much, much higher for the uncorrected data compared to the corrected data by accounting for the changes in optical properties. And when you actually looked at the difference in flow between at the 100 millimeters of mercury occlusion, this difference between uncorrected and corrected was actually a 200% difference. So one of the takeaways for us from this was that if you're going to be having these large changes in optical properties, especially if they're attributed to blood volume, it's really important to take into account the optical property changes.

That's the, that's the. So the question is, these oscillations, what are they? Right? So that's actually due to each contraction of the heart. So if you were to zoom in here, it doesn't look like noise, it's actually each heart rate. Yeah. Um, basically, it's the pulsatile waveform that's in the finger at that time point. Yes. Yes. So the, the question is, do those oscillations, yeah, yeah, that's a good question, Boson. So the question is, do those, kind of the amplitude of the oscillations, uh, in the SPG give any information about the blood pressure? Right now, I don't know the answer. Um, some of this is a little bit counterintuitive, at least in the way that I'm thinking, because as I feel like you would increase your pressure that's on your brachial artery, in some ways you're kind of maybe decreasing or increasing, it's hard for me to really know. I can see how you're kind of constricting the blood vessels and therefore you would get a larger amplitude because there's more pressure on the walls that you're imaging, or it can be decreased because if someone has a low enough pressure, then it would actually maybe be a lower blood pressure. Um, so the, right, right, right. So I, sure, basically looking at this amplitude, right? Yeah. So, so that's where for me, originally when I was going into this, I thought at 100 millimeters of mercury, basically on the arm cuff, I'd actually originally thought the amplitude would be smaller because you're, you know, you're cutting off some flow. But maybe that's not the case. Um, and there could be, you know, there are techniques that basically, um, on the finger to get non-invasive blood pressure where they actually will tighten up, um, and they're using that information to get their blood pressure. It's basically like a small, um, blood pressure cuff on the, um, so that that could effectively be what this is. Yes. Um, I think it definitely warrants further investigation, um, but right now I don't know the answer to that, but I think it's definitely something interesting to look at for sure.

So, so this is just from the finger, right here. So it's, it's a small area. So that this would be more, you know, being able to take it to the patient, put this on, and let it go, kind of. Um, so looking at a very small area. But within, you know, this whole device, you have your, uh, your coherent light source, your camera, your lens, and everything. But looking at a very, very small area. But yeah, this is, uh, was definitely a kind of a proof of concept for the coherent spatial imaging technology. Um, but I think a lot of work can be done potentially utilizing this device to get blood pressure information or some other, uh, different analytes as well. Yes.

Sorry, can you repeat that? So basically going from 100 to zero. Yeah. So that's a, basically if this is the occlusion pressure, yeah, it's basically the cuff's going from 100 to zero. So effectively, the, there's no like true hyperemic response for the corrected blood flow, which I'm not 100% sure you would expect because you're not cutting off the arterial side. The arterial side is continuously going in. Um, but for the uncorrected, it does appear that there's a, uh, increase or hyperemic response. But it doesn't seem to at least stabilize within this one-minute period, where this one, it appears stabilized. But, you know, maybe the uncorrected over time will kind of trend down more closely to the baseline period. Simultaneously, no. Um, we, this sum, we've actually been working on increasing the number of devices that we have. Um, so we're in the validation stage. Probably next week, we'll actually be starting to run these experiments for increasing the number of devices we have. This is N of one device, but we're, uh, hopefully by the end of summer, we'll have five devices, is kind of what we're going towards. Yeah. Yeah.

That, that's a good question. Um, most, most likely what happened was that the subject during that period took a deep inhale. And actually, what happens when a subject takes a deep inhale is that your lungs expand, and it's actually putting some pressure on your heart, where you actually will get a decrease in blood flow. Um, that, that is typically what would happen in something such a large change like that. It wasn't observed in every subject we measured. Um, so basically, the, the question was, during this 80 millimeters of mercury, why do you have this large decrease in the flow? And my guess is that the subject took a deep inhale because we didn't see it in any other subject. But maybe there's something more there physiologically in that subject that we don't really know. Like Vava? Yeah, yeah. We, we've done a little bit of that work as well, um, with the Michelle K's group, um, but yeah, for this one, it was more like, ideally, the person would just have normal breathing the entire time. You know, this is a relatively long acquisition, you know, eight minutes long, so it's kind of natural for you to be kind of like, you know. So no, no fault to the subject. Yes.

Yeah. So, um, I believe Bernard will be going more into that into the next lecture, a little bit about the impacts of how spectral contrast is impacted by absorption and scattering. But there are basically models where where you input optical properties to get your corrected flow value. So I'll let him kind of go over that a little bit more in the LA. Also, there's a little bit of that, um, so it's both. So basically, um, if we were to look at, say, one of these line profiles, we get our, uh, LSI, the, the SPG, looking at the speckle contrast. But if we actually use kind of the banana path of all the photons that are acquired by the camera, U, that's the diffuse reflectance spectroscopy, where we can then fit, uh, basically we have a Monte Carlo simulation, then we fit basically the source-detector separation and have a lookup table to try and match the physical data to the simulated data and extract the optical properties that best match the simulated data. And then from there, we then input that into kind of correcting the blood flow. That's correct. Yeah. So, so basically, I should have included an image. If you were to have an image of basically the reflectance as a function of space, it's brighter and then it decreases the intensity as you, uh, get further away from the source. Okay, that's till 10, right? Or do we want to cut it a little earlier? Okay, okay.

So I'll just talk about the last. This is the last application I have, which was a lot of my PhD work that I did with, uh, Yama Abari, which was a really, really eye-opening experience. So I think it's probably you guys had some of this, uh, when you, uh, met with Yama and you gave a little talk. But just going to briefly go over what cardiac arrest is. So hopefully everyone here has a normal beating heart here, um, outputting blood to the rest of our body. Something may trigger cardiac arrest to occur where our heart doesn't output blood anymore. So that's really what cardiac arrest is. It's different from a heart attack. It's when you're not outputting blood from your heart to go to the rest of your organs anymore. Now, if someone doesn't have blood flow going to the rest of their body, one of the things that you need to do is try and, uh, get them alive. So to do that, start doing cardiopulmonary resuscitation, inject different drugs as needed, chest compressions are really, really important, sometimes defibrillator, depending on the type of cardiac arrest. And if everything kind of goes well, then the heart can regain its function. It's still a really, really low, uh, success rate. Some of these statistics may be a little bit outdated, but typically it is difficult for someone to regain their heart function after having cardiac arrest. But another organ, which is what we really focused on, was looking at the brain, where actually a really, really small percentage of individuals who have cardiac arrest have good neurological outcome. And what good neurological outcome means in this context is be able to go about do their daily, uh, functions without needing a constant caregiver with them. So they can have some assistance needed, but, uh, not like full-time care where someone always needs to be with them.

So to, to do this, um, we kind of have the current standard of clinical care, um, which is really looking at three things: the arterial blood gas measurements, so taking blood out, looking at the different components of the blood, your pH, uh, your bicarbonate. Then you have invasive arterial blood pressure, so having someone canulated and be able to look at the blood pressure over time. And then at least when looking at the brain, using EEG to look at that brain electrical activity. So with, uh, what Dr. Abari did, he basically made this animal ICU here, at UCI, where you have those current standards of clinical care. You have electrodes that are able to monitor the brain electrical activity. You have the invasive blood pressure, uh, from the canulated from artery with the blood pressure transducer, and then be able to withdraw blood and look at those blood components. But one of the things with some of the limitations is that it's invasive. So you need to either draw blood or you need to perform a canulation. It provides indirect brain monitoring, so inferring what's happening in the blood or from the, uh, blood pressure to see, hey, this is what's happening in the brain. Or if you do measure the brain, it's really only looking at electrophysiology, but not necessarily looking at hemodynamics. So in our collaboration, uh, we added an LSI arm to be able to look at blood flow, uh, in a rodent model where craniectomy was performed. So the skull was removed, in this data, to be able to resolve the blood flow dynamics that are occurring in the brain.

So I'm going to walk through, uh, kind of an example experiment. Um, what this experiment has is that there is a five-minute baseline period, that is followed by a cardiac arrest period. And the way we induce cardiac arrest is through asphyxiation. We basically, uh, inject a paralytic, so the rodent can't contract its diaphragm when breathing. And, um, to start, the rodent is on a ventilator, so we're controlling its respiration. As soon as we turn it off, it asphyxiates and induces cardiac arrest. So no more blood out coming from the heart. Then, uh, Dr. Abari, uh, and, you know, the first time I saw this, I was like, wow, my mind is blown. But basically, he would actually do chest compressions on the rat, uh, during the experiment. They were giving epinephrine, sodium bicarbonate. And then if everything kind of goes well, then the rodent would become resuscitated, and we would look at the recovery in terms of the different parameters for, uh, the first hour or two.

So starting with, kind of, the baseline dynamics, we have normal blood pressure and blood flow. And all these spikes in the EEG, is the brain firing effectively. Then, once we turn off the ventilator, the blood flow rapidly declines. So you have an absence of blood pressure. You can see the speckle flow index, uh, where the blood flow is basically all blue, so no blood flow. And then the brain, because it's not getting any nutrients, in order to fire, the brain electrical activity is gone. So then, you know, the drugs are administered, Dr. Aquari would start CPR. The, uh, rodent came, uh, was able to, uh, be brought back from the dead. Um, and the blood pressure was restored. You had this hyperemic response of blood flow that's going into the brain. However, during this period, even though you have blood flow going to the body, the brain, uh, electrically, is still not firing anymore. So as the experiment goes on a little bit further, uh, the blood pressure, or the blood flow, would decrease and then stabilize in a hypoperfused state. The blood pressure will kind of slowly go down and maintain roughly, uh, equal, uh, values to its baseline levels. And then the brain will actually start firing again at some point later. And for a lot of my PhD work, one thing that we were looking at was when there are these rapid hemodynamic changes, uh, that are occurring when the brain is not, uh, active yet.

So to, to do this, I want to start focusing on the different phases of the EEG. So, um, starting with the asphyxia period, the EEG is completely silent within about 30 seconds, so no neurons are firing. These are typically noise that is occurring. ROS is the resuscitation time point where the, the brain still is not electrically active. And then at some point, the brain starts firing again, and that's what we call bursting, so brief periods of electrical activity followed by electrical silence and electrical activity again. So now that we know, kind of, what EEG looks like, how does it look like in respect to the cerebral blood flow? So we have the hyperemia phase when there's no brain electrical activity. And one thing that we observed, just when kind of looking through all the data, was that at the, when this the LSI data has a decreasing blood flow being shown, that typically is around the same time that the initial burst or the start of electrical activity is beginning. So we wanted to look at this a little bit more. And the reason why we did is because other papers have shown, without any optical data, but from an electrical, for an electrical activity perspective, is that the earlier and more frequent this burst is, it typically results in better neurological outcome. So we wanted to see, hey, does, um, this cerebral blood flow that we're able to measure, kind of play a role in the initial EEG burst that occurs?

So to do this, we wanted to see, hey, is there a threshold of total perfusion from resuscitation to when that burst occurs? So what we did was using the relative blood flow signal, we did the area under the curve to look at the, basically the total amount of, uh, blood that the brain received from the moment of resuscitation to when that burst occurred. So basically, the integral from ROS to the burst of the, uh, blood flow signal. And what we did see was that the longer the time from resuscitation to the starting of brain electrical activity, more perfusion was needed. And if we increased the amount of time of when the ventilator was off, so longer period of cardiac arrest, um, also more perfusion was needed. So, and going back to, hey, does total perfusion kind of guide or is it the reason for the starting of brain electrical activity? Well, by just looking at total perfusion, it wasn't so. But we do see that is kind of this, kind of trend towards increased, uh, perfusion needed for more severe cardiac arrest.

So what we did was then, okay, so can we just normalize the total perfusion by the amount of time that the subject doesn't have blood flow going to their brain? And we call this the predictive burst ratio, for the term. And once we did that, and we looked at both blood flow and mean arterial pressure for this, is that regardless of the severity of cardiac arrest, especially for the blood flow, there was no difference in this predictive burst ratio, just by normalizing it to the asphyxial duration. And that was very, very similar with the mean arterial pressure. So now that we have this predictive burst ratio, can we use this to try and predict, uh, when the brain would restart from cardiac arrest? So to do this, we, uh, had a, uh, uh, the blood flow on the y-axis and the time after resuscitation on the x-axis. And we're going to be looking at the predictive burst ratio as a function of time. So what we have on the right is the predictive burst ratio as a function of time. And we're using basically all the predictive burst ratios from all the experiments we had to see if we can predict it. And we did is we fit a line to that predictive burst ratio that was five minutes in length. So starting five minutes, we're like, hey, right now the prediction says it's going to be 34 minutes after resuscitation. But we wouldn't actually accept that because this R-squared value is not a good enough fit for what we were doing. So we would say that the R-squared would need to be greater than 0.98, comparing the actual data, the blue circles, to the red line. So the experiment would keep on going, and you can see that the R-squared is getting higher and higher. And as some point, it reaches that 0.98 threshold. So seven minutes after resuscitation, we're our fit is good enough to say, hey, we're confident in the prediction that it's going to be about 20 minutes. So you can see that the prediction kind of went lower and lower as our fit got better. And now we actually see what the actual, uh, burst time was, or the resumption of EEG activity was. And the actual one was about a minute off from the predicted, uh, time. So using blood flow, it actually could predict the burst time, um, within a reasonable degree, um, on average, it had a less than 10% air, um, for both five and seven minutes. And then just looking at mean arterial pressure, it didn't do too bad, but it did have a larger air than just blood flow alone.

So kind of the, the key takeaway from, uh, this initial work was that, oh, sorry. So can you repeat your question? Oh, the area under the curve. So without the correction, um, so the predictive burst ratio between five and seven, we didn't see any difference. So this is one experiment. So basically, this was a five-minute, uh, cardiac, actually, this was a seven-minute cardiac arrest experiment. So at the five-minute period, it said the predicted time was 34 minutes, is how long after resuscitation the brain would start firing again. But the R-squared of that linear fit to the actual data was only 0.91. So we had a threshold that, hey, the fit needs to be at least 0.98, or we're going to reject that prediction time. So we can, we pulled five and seven minutes together. So this was, this was used for both five and seven minutes. So the reason why I think it worked, I don't know for sure, but we normalized to the asphyxial duration. So that was kind of one of the terms in, uh, this predictive burst ratio, that normalization. Um, and I mean, it seemed to, this was kind of the, the second iteration. This actually isn't in what we published because I modified it later that got more accurate by having kind of a dynamic fitting method as opposed to just being like the same bounds effectively all the time. Um, so yeah, it's actually for five and seven minutes, this this approach.

So, so effectively, I'm going to keep on going here. So this beginning portion, it's a little bit nonlinear. And this was more apparent the more severe the cardiac arrest was. But then at a certain point, it actually becomes quite linear over time. And this is kind of the ideal portion that you would want to fit to be very accurate because as it's going on, it would be a very, very linear line of that predictive burst ratio over time. And it was that way for five minutes and seven minutes. We didn't do like more severe, like eight minutes or, uh, less severe either. Um, the, we did try eight minutes actually to start with, but our mortality was too high that we couldn't actually do any of this stuff very well. Um, but yeah, so, so it's effectively trying to get when the predicted burst ratio as a function of time is linear to get that predicted resumption of electrical activity.

So with that, going into kind of the take-home messages is that LSI, fairly simple technology, just laser shine on your tissue and collect it with the camera and have a, a CPU to control it. Um, it can be used in clinical devices as well as preclinical, um, for surgery perspectives, diagnostic perspectives, um, and it can be in different form factors. So, you know, I talked about the clinical-based devices on the cart, going into the handheld, going into the finger clip. Um, so having it from a non-contact as well as a wearable perspective. And with that, if there's any more questions, I would gladly take them. Yes.

Yes. So a lot of it goes into this. Yeah. So, so the biggest thing is most likely the software. So the actual off-the-shelf components are fairly straightforward, but it's the software that they spend, you know, time and effort trying to get. They, you know, try and calibrate their data, which we do as well. It's just a, you know, it's all the, the R&D that goes into creating that device where, you know, we can make the same thing in the lab for under 5K, um, and be a pretty good device. But, you know, all the things that go into it, um, with having a commercial product, you know, getting approvals for, um, all that, but also as a financial burden, I believe that they had a liquid-based phantom for the commercial systems, that they would, uh, get. Yes.

Yes. So, so this one, a lot of times would just be artifact because, um, the rat was just resuscitated, they're trying to, you know, have all the, the tubing and the drugs and the mouse could be moving. Um, that's why sometimes there's artifacts that are picked up on the EEG. Um, but, um, sometimes there are. So this EEG data was from 1 to 150 Hertz, um, but there can be slow frequency oscillations like in the delta band. Um, so from this, so at least I don't know if they're doing it now, but towards the end of my PhD, we're using, uh, DC instead of, so the EEG is an AC amplifier that's used where you can look at a DC amplifier, and that one would target the slow waves a lot better. This one, we would actually, there would be a hard filter in the preamplifier that would kind of reject any really low frequency data. We tried reconstructing it a little bit, but it wasn't, uh, too successful. Um, so I think that there are probably low frequency dynamics that occur there, but at least with this data, we weren't able to really resolve it.

So the question is, what is the wavelength and the optical power? So the big variability, um, so the wavelength that we typically use are the red to near infrared. So a lot of our work has spanned from like 633 nanometers to 850 nanometers. So those would be the wavelengths that we typically use, kind of going into that optical window where the absorption is much lower. Um, and then in power, uh, there are safety considerations that you need to target, um, typically more the eye safety than the skin safety because it's a fairly diffuse sample that we're illuminating. Some of the finger clip devices, you can get away with just a few milliwatts. If you're trying to do really long distances, for example, then you need a higher power laser. So we would typically go to like 100 or 150, but our beam is coming out really, really fast that it's a fairly large field of view, um, with a low irradiance over the entire field of view.

So the question is, what's the maximum field of view? The, the biggest field of view that I've probably worked on is about 20 by 20 cm. Um, and in the smallest is, uh, yeah, so that, that's where you do need to consider like optical power and everything because you do need enough signal for your data to actually be able to look at these flow dynamics that are occurring. Um, and then for those where you do have a large field of view, I definitely do look at like eye safety calculations because I want to maximize the light that I send onto my sample, but still be able to be eye safe for, for people, especially if, uh, uh, say eye protection is not being used in those applications.

So the, the question is, how would we compare this with OA? We've done a little bit. Um, it's a, I think it's a slightly different. I know it's still based on the interference, but, but, you know, some of the results make sense and others don't. Um, and I don't think I've done enough to really comment fully, but we have them a little bit, and some of the results, like I said, make sense and others I'm like, this doesn't make sense at all. Like, I don't know which is right or or anything like that. Yeah. At least for the OA part, I think the visualization of vessels is good, but the actual flow information sometimes there's not, and I don't know how to merge or make them co-exist well.

So the question is, how have I done any work on tumor angiogenesis? I personally have not, but in the last, I know that they did do some tumor imaging at one point. I don't know if that's been published or anything. No, okay. It's been used, but I guess not, uh, uh, published.