Understanding PPG Sensors: The Math & Science of Smartwatch Health Tracking

Everything we do is available on buyingkills.com. So you can find all the tools on the landing page. Go check them out. Provide your feedback. We'll jump into this pulse vision. It's a PPG simulation. Yeah, it has the explanation at bottom of the page. You can now use full screen. And full screen can bring me in the controller. So you can control parameters and things. You can go try it. Try it yourself. The main parameter here is to do with the dicrotic notch intensity. So in PPG you have this dicrotic notch. Anyhow, we now also on the page have this demo. Now the demo is linked to this 22-minute interactive audio audio guide. So essentially once we play the audio guide, you also have the the page moving the controls and things showing you around. So see how that goes. >> Have you ever taken off your smart watch in a dark room, turned it over, and just um stared at those little flashing lights on the back? >> Oh, yeah. Usually it's that frantic blinking green, right? Or maybe a red glow if it's a newer model. >> Yeah, exactly. And if you've ever paused and wondered how a tiny, honestly cheap LED is somehow looking through your skin and telling you your heart rate, your oxygen saturation, and your stress levels, well, you are not alone. >> Right. On the surface, you're just looking at a fancy flashlight. >> Yeah, but underneath there is an entire invisible world of biological data being captured. >> It really does feel like science fiction when you stop taking it for granted. I mean, you're shining a light onto a wrist, and suddenly a computer chip knows exactly what the left ventricle of your heart is doing. >> It's wild. >> Right. But getting from that flash of light to a readable heart rate, it isn't magic. It's a highly precise, incredibly complex optical calculation that most of us, well, we just never think twice about it. >> Which is exactly what we are going to explore today. Okay, let's unpack this. We have a really fascinating source document today for our deep dive on something called Pulse Vision. >> Yes. >> Now, before your eyes glaze over at the terminology, Pulse Vision is essentially a digital sandbox. >> Right. Yeah, that's a good way to put it. It's a simulator created by a company called Bionic Chaos. >> Right. And it basically lets biomedical engineers peer under the hood of this technology. It allows them to see exactly what those smartwatch lights are seeing, um, manipulate the variables, and decode the rhythmic language of the human cardiovascular system. >> Exactly. >> So, our mission today is to dig into the hidden biomechanics and the surprisingly beautiful math that makes wearable health tracking possible. >> And it serves the perfect lens for this topic. Pulse Vision takes something entirely invisible to the naked eye, you know, the microscopic rhythmic expansion of your blood vessels, and it translates it into numbers and physical ways we could actually dissect and understand. >> Okay, so let's start with that physical glowing light. I know there is a massively complicated medical term for what that light is doing when it hits your skin. >> There is, yeah. The technical term is photoplethysmography. >> Wow. >> Yeah. It's a mouthful. >> Oh. >> So, everyone in the field just calls it PPG. >> PPG, got it. >> Right. And if you break down the Greek roots, photo means light, plethysmo means increase in volume, and -ography is to write or record. >> Oh, okay. >> So, we are quite literally using light to record changes in blood volume inside an organ or a tissue bed. >> Wait, let me wrap my head around that. How does shining a light tell a sensor anything about volume? >> Well, it comes down to absorption. Think about what happens when your heart beats. Your left ventricle contracts and violently pushes a surge of blood out into your body. >> Okay. >> This creates a synchronous arterial pressure wave. >> Mhm. >> And as that wave of blood reaches your peripheral microvascular tissues >> Like your fingertips or earlobes. >> Exactly. Or your wrist under a smart watch. Those tiny blood vessels physically expand to accommodate the extra blood. >> Okay, so they swell up for split second. >> They do. And blood is very good at absorbing certain wavelengths of light. Your smart watch basically has two main components on the back. An emitter that shines the light and a photo detector that catches whatever light bounces back. >> Makes sense. >> So when your blood vessels expand with a heartbeat, there is suddenly more blood in that localized area of tissue. More blood absorbs more light. Therefore, less light bounces back to the detector. >> Oh, wow. >> Yeah, and when the vessel relaxes and the blood drains out, less light is absorbed and more light hits the detector. >> So by tracking that tiny rapid fluctuation in light intensity, the sensor is actually tracking the physical pulsing of your arteries. >> Exactly. >> That is incredibly elegant. I mean, more blood equals less light bouncing back. >> Right. >> But the data they get from bouncing this light around isn't just one flat metric. I was looking through the source material and the resulting signal actually has two distinct parts. >> Yes, it does. >> I want to try an analogy here to make sense of the physics. Tell me if I'm off base, but I picture the signal coming back to the sensor like an ocean. >> Okay, let's see where you're going with this. >> So if you're looking at the ocean, you have the deep, heavy, constant body of water below. >> Right. >> In our light signal, this seems to correspond to the DC offset. It's the steady, non-pulsatile part of the reading. >> Yes. >> It represents everything in your wrist that is just sitting there, stationary. Your bone, your muscle tissue, and your baseline venous blood. It's the deep water that doesn't change from second to second. >> Exactly. >> But up on the surface, you have waves continuously crashing. That's the AC pulsatile component. Those surface waves are the actual surges of arterial blood driven by the heartbeat constantly rolling in over the deep water. >> That is a brilliant way to visualize it. >> Yeah. >> The photodetector is forced to look at the entire ocean. The deep water and the waves together. >> Right. >> But it has to mathematically separate them. It really only cares about measuring those crashing surface waves against the static depth of the water below. >> Okay, that makes sense. >> And what's fascinating here is how we extract oxygen saturation or SpO2 from this data once it isolates those waves. We can calculate something far more complex than just heart rate. >> Now, this is where I get lost. I understand counting the crashing waves to get a pulse, but how on earth can a single flashing light tell you the exact percentage of oxygen floating around inside the blood making those waves? >> Well, to pull that off, we have to lean into physics. Specifically something called the modified Beer-Lambert law. >> The modified Beer-Lambert >> Right. This law dictates how light absorption changes based on the physical properties and concentration of a substance. Clinical pulse oximeters and some high-end wearables, they don't just use one light. >> Oh, they don't? >> No, they use two specific wavelengths. They use 660 nm visible red light and 940 nm infrared light, which is totally invisible to your eye. >> Why those two specific colors though? Does oxygen react differently to red versus invisible infrared? >> It does because it physically changes the hemoglobin. Hemoglobin is the protein in your red blood cells that carries oxygen. When hemoglobin binds to oxygen, its physical structure actually changes shape. And that shape change alters its color and how it interacts with light. >> Oh, wow. >> Yeah, so oxygen-rich blood absorbs much more of that invisible infrared light and it lets the visible red light pass right through it or bounce back. >> Ah, so that's why highly oxygenated blood looks bright red. It's rejecting the red light. >> Exactly. But deoxygenated blood behaves the opposite way. It absorbs more of the visible red light and lets the infrared pass through. >> Oh, I see. >> So the smart watch is rapidly flashing both red and infrared lights, one after the other, literally hundreds of times a second. It separates the crashing waves from the deep water for the red light, and it does the exact same for the infrared light. >> Okay. >> And it compares the two, calculates something called the ratio of ratios, which engineers just call R. >> So it's a massive comparative math problem happening under the skin. >> It is. >> Yeah. >> And once the sensor determines that R variable, the pulse vision simulator reveals the surprisingly simple middle school algebra that finishes the job. >> Wait, middle school algebra? >> Yeah, the algorithm uses a linear equation. SpO2 = 110 - 25 * R. Plug your ratio into that, and boom, your smart watch tells you your blood oxygen is happily sitting at 98%. >> It is wild that such a simple equation is the final gatekeeper for a metric that might tell you to rush to the emergency room. >> It really is. >> So we have this ocean of data coming in, we've separated the deep water and counted the crashing waves, but I imagine these waves aren't just perfect smooth arcs. Like when you look at the simulator's readout, it's not a gentle rolling hill. >> Not at all. The morphology, you know, the actual physical topography of the pulse wave, it has distinct features, and those features hold incredible amounts of health data. >> Like what? >> Well, the first thing you'll notice on a PPG waveform is a sharp, sudden upward spike. That is the systolic peak. It represents the maximum expansion of your blood vessel during the strongest part of the heartbeat. >> Okay. >> In the pulse vision simulator, they give this an artificial sharpness factor of five just to accurately replicate how violently that left ventricle throws blood into your system. >> Right, the initial spike makes sense. But what goes up must come down. The downward slope, um, the catacrotic limb, I believe the source calls it, seems to have a lot going on. >> It does. The descending slope is arguably the most important part of the wave. Halfway down that catacrotic limb, there is a very distinct, sudden, little dip and secondary bump. It's called the dicrotic notch. >> See, I was looking at that in the simulator documents and here's where it gets really interesting cuz I have to challenge this a bit. >> Okay. >> In Pulse Vision, they have all these specific sliders. You can change the dicrotic notch intensity and you can adjust the dicrotic notch phase offset, which is set by default to 0.42 on their scale. >> Right. >> You can artificially make this little bump bigger or slide it around, but looking at the visual of the wave, it's just a microscopic hiccup. It's a tiny blip on a downward curve. >> Yeah. >> Does that actually mean anything biologically or is it just an obsession for the engineers building the software? >> Oh, if we connect this to the bigger picture, it is profoundly significant biologically. That tiny blip is not noise and it's not a glitch. The dicrotic notch corresponds to the exact microsecond your aortic valve snaps shut. >> Wait, really? A valve closing in my chest creates a visible bump on a light sensor on my wrist. >> Yes, through what's called a retrograde arterial wave. >> Okay. >> When your heart pumps, it pushes blood through the open aortic valve, but when that pump finishes, the valve snaps shut to prevent the blood from flowing backward into the heart. >> Right. >> When the forward-moving pressure wave hits that freshly closed valve, it bounces off it. It creates a secondary, smaller echo of a wave that travels all the way down your arm and out to your extremities. That echo is the dicrotic notch. >> That is fascinating, but why do we care how big that echo is? >> Because the prominence of that notch is a direct indicator of your vascular compliance. >> Yeah. >> It tells us how stretchy and elastic your arteries are. >> Oh, I see. >> Think of a healthy artery like a brand new rubber band. When the heart pumps, it expands easily and when the valve shuts, it snaps back with excellent elastic recoil. That snappy recoil preserves the echo of the valve closing, resulting in a very sharp, prominent dicrotic notch on the waveform. >> Okay, so what happens if it's not a new rubber band? >> Well, as we age, or if someone develops cardiovascular disease, their arteries stiffen. >> Mhm. >> They become less like rubber bands and more like stiff garden hoses. >> Yikes. >> They lose that elastic recoil. When a pressure wave bounces off the valve in a stiff artery, the energy just dissipates. The dicrotic notch starts to flatten out, smooth over, or disappear entirely from the reading. >> Wow. >> So, by analyzing the shape of that one tiny blip, an algorithm can effectively estimate the biological age and stiffness of your arteries. >> So, a doctor could literally look at the raw data from my smart watch and say, "Your heart is beating at 70, but your arteries look like they belong to an 80-year-old." Which I guess is exactly why a sandbox like Pulse Vision needs to exist. You can't just study healthy 20-year-olds if you're trying to build a life-saving wearable. >> Precisely. You need to see how the waveform degrades in extreme conditions. The source material outlines how Pulse Vision allows researchers to instantly jump between physiological states. >> Right. >> You can start with a normal resting preset. So, that's 70 beats per minute, 98% SpO2. Then, with a click, you can model exercise tachycardia, pushing the simulated heart rate to 140 with a slight oxygen dip to 96%. >> Or they even have an athletic condition preset, right down to 48 beats per minute and 99% oxygen. >> Yeah, exactly. >> And then there are the dangerous states. They have a hypoxic desaturation preset. This drops the heart rate to 55 beats per minute and crashes the oxygen saturation down to a critical 86%. >> Right. >> When I was reviewing the telemetry workflow for that specific state, something really stood out. When the oxygen drops that low in the simulation, the overall physical height of the AC signal, like the literal size of the waveform on the screen shrinks dramatically. >> Yeah, and that ties perfectly back to what we discussed about the modified Beer-Lambert law and light absorption. Remember, the sensor relies on the contrast between the highly oxygenated arterial blood and the static baseline tissue. >> Right, the color change of the hemoglobin. >> Exactly. When a patient becomes hypoxic and their oxygen saturation plummets, their arterial blood starts looking a lot more like their baseline venous blood. The stark optical contrast vanishes. >> Oh, because there is less oxygenated blood to block the light? >> Yes. The peaks of those crashing waves physically shrink on the sensor's readout. >> Right. >> It's a massive challenge for the hardware to keep tracking a signal that is vanishing into the background. >> The simulator actually has an audio feature that drives this home, right? It's called pulse audio synthesis. >> Oh, yeah. That feature's brilliant. >> It generates an acoustic click for every heartbeat, much like a hospital monitor, but the pitch of the click dynamically scales with the oxygen saturation. >> Yeah. >> So, as the SpO2 crashes from 98% down to 86%, the pitch of the beeping fundamentally drops, giving this really ominous auditory biofeedback. >> It's an incredibly effective way to intuitively understand the patient's deterioration without even looking at the screen. And speaking of the signal shrinking, we have to touch on another crucial variable in the simulator. The perfusion index, or PI. >> The perfusion index, let's see. The formula here is the AC component divided by the DC component multiplied by 100 to get a percentage. >> Right. >> If I'm translating that back to our ocean analogy, um it's calculating exactly how big the crashing waves are relative to the total depth of the deep water. >> That's spot on, yeah. PI measures peripheral perfusion, which is just a fancy way of saying how much blood flow is actually making it all the way out to the extreme edges of your body, like your fingers and wrists. >> Okay. >> A high PI, say above 5.0%, means your blood vessels are wide open, fully vasodilated, and blood is surging through freely. But if the PI drops below 0.5% it indicates severe peripheral vasoconstriction. Your vessels have clamped down tight. >> So, what does this all mean for you in your daily life? Imagine you go for a run in the middle of January. It's freezing outside and within 10 minutes your hands feel like blocks of ice. >> Oh, definitely. >> Biologically, your body is restricting blood flow to your arms and legs to keep your core organs warm. Your vessels are clamping down. That means your perfusion index is plummeting below 0.5%. >> Right. >> The crashing waves in your wrist have become tiny little ripples and suddenly your brand new top-of-the-line smart watch loses your heart rate. It just can't find the pulse anymore. >> And that scenario is the ultimate nightmare for biomedical engineers. Cuz if you're running in the cold, your PI isn't just low. You're also swinging your arms, shivering, and sweating. >> Yeah. >> That introduces what the source calls stochastic motion artifacts and optical sensor baseline drift. >> Meaning random, unpredictable noise destroying the signal. >> Completely destroying it. Every time your watch slides a millimeter across your sweaty skin, the underlying baseline tissue changes. The depth of our simulated ocean instantly shifts. So you have a microscopic shrinking pulse wave buried underneath massive spikes of random light variation caused by your arm swinging. The sensor has to mathematically hunt through all that chaotic noise just to find the rhythm of your heart. >> Which brings up a massive question. How does it do that? How does the watch, or rather how does the Pulse Oximetry simulator mathematically reconstruct a perfect biological wave out of all that chaos? >> It's complex, for sure. >> The source material gets pretty dense here, mentioning that the simulation operates inside an HTML5 canvas using a dual Gaussian sum model. >> It sounds incredibly intimidating, I know. But let's demystify it. A Gaussian curve is just a mathematical term for a classic bell curve, a rounded hill. >> Okay, a bell curve. >> We know that a human pulse wave isn't just one smooth hill, right? It has the primary systolic peak, and then it has that secondary echo, the dicrotic notch. >> Right, the two bumps. >> So, instead of trying to draw one weird, complicated shape, the math engine uses two distinct bell curves. >> Ah, dual Gaussian sum. >> Exactly. It uses a large, sharp bell curve to simulate the massive left ventricular pump, placing it early in the phase cycle, like around .2 arrow. Then, it mathematically adds a second, smaller, wider bell curve further down the timeline, around .42, to represent the dicrotic notch. >> That is so clever. >> By summing these two curves together, it perfectly mimics the complex topography of a human pulse. >> The source also points out a really cool trick they use. It says the math wraps the distance periodically across cycle boundaries to eliminate step discontinuities. >> Yeah. >> So, it ensures that the end of one heartbeat flows perfectly seamlessly into the beginning of the next without any jagged cliffs or drops in the data. It's endlessly plotting these dual bell curves, calculating the heart rate variability, updating everything in real time. >> The mathematical elegance is staggering. And what's really exciting are the future directions Bionica Chaos is planning for this engine. Right now, Pulse Engine heavily focuses on red and infrared light. >> Right. >> But, they're preparing to integrate green optical sensors at 525 nanometers for wrist wearables. >> Why green? Most smartwatches I've seen use green lights on the back, now that I think about it. >> Because green light doesn't penetrate as deeply into the tissue as red or infrared light. It mostly scatters in the superficial capillary beds just under the skin. >> Oh, I see. >> It turns out that makes green light far less susceptible to all those motion artifacts we talked about. If you're swinging your arm, the deeper tissues and muscles are shifting wildly, which wreaks havoc on infrared light. >> That makes total sense. >> Right. But, the shallow capillary beds stay relatively stable, making green light the gold standard for fitness trackers where the user is constantly moving. >> Okay, that makes so much sense. But, they aren't just changing light colors. The source also notes they're going to model clinical cardiovascular pathologies. They want to simulate ectopic beats, which are essentially skipped or extra heartbeats. >> Yeah. >> They are building models for premature ventricular contractions and even atrial fibrillation or AFib, >> You what? >> upper chambers of the heart beat completely out of coordination. But, let me ask you this. Why bother? Why build immensely complex dual Gaussian math equations to simulate AFib? Wouldn't it be vastly easier to just pull real hospital data? Just record a patient having AFib and feed that recording into the simulator. >> This raises an important question, but it gets to the core of why synthetic data is revolutionizing medical engineering. >> How so? >> Think about a real patient recording. It is entirely static. If you record 10 seconds of someone experiencing AFib, you have exactly those 10 seconds. >> Right. >> You cannot dynamically alter their arterial stiffness on the fly. You cannot ask their blood oxygen to smoothly drop to 86% while keeping their heart rate exactly at 115 beats per minute. >> Oh, wow. I didn't think of that. >> And you certainly cannot isolate the specific optical noise of a smart watch sliding precisely 2 mm to the left during a contraction. >> I see. So, if a software engineer's trying to train a watch to detect AFib, they don't just need one recording of a heart attack, they need to know what a heart attack looks like in a 20-year-old with stretchy arteries, an 80-year-old with stiff arteries, someone running in the cold, someone sweating in a sauna. >> Exactly. Synthetic data gives you god-level control over the variables. You can inject very specific amounts of baseline drift or motion artifact to see the exact breaking point of your algorithm. >> That is incredible. >> You need a mathematically perfect sandbox to rigorously stress test the software before you ever trust it on a human wrist. >> And the fact that the sandbox is largely open access is just fantastic. The source mentions the educational module is released under a Creative Commons open access license CC-BY-NC 4.0 for non-commercial research, academic study, and clinical education. >> Yeah, it's a huge resource. >> Now, Dr. Yuri Benno do handle commercial licensing for a proprietary hardware through Bionic Cloud, but they've built this entire ecosystem. They have the Cardio Quest environment for ECG training, a pacemaker electrophysiology model. It's like a fully digital medical wing. >> It really highlights the convergence of human anatomy, advanced physics, and computer science. >> It absolutely does. So, the next time you take off your wearable and you see those little green lights flashing against your nightstand, you'll know what's really happening. You aren't just looking at a fancy LED. You are looking at a complex optical instrument rapidly applying the modified Beer-Lambert law. It is calculating ratios of red and infrared light absorption, mathematically fighting through the noise of your movements, and hunting for microscopic dichrotic notches to estimate the biological age of your arteries, all in real time. >> It makes you appreciate the engineering on your wrist a whole lot more. >> It definitely does. And there is one final provocative thought from the source document's future directions that I want to leave you with. Bionic-Project plans to eventually link this Pulse Vision simulator with an ECG simulator to estimate continuous non-invasive blood pressure. >> Oh, yeah. NIBP. >> Right. They want to use something called pulse arrival time or PAT. Essentially, they want to measure the exact millisecond delay between the electrical signal of your heart firing in your chest and the physical surge of blood arriving at your wrist. >> amazing concept. >> Right. If combining a simple light sensor with an electrical timing mechanism can eventually track your blood pressure perfectly 24/7 without ever needing a cuff squeezing your arm, what other invisible metrics are our microvascular rhythms broadcasting right now every single second that we simply haven't built the math to decode yet. >> Okay, that was notebook LM overview that is available on the page as well. So, when you go play it, it will actually scroll through the parameters essentially show you around what's going on. I'm not actually 100% sure I did I did a it did a perfect job, but uh just the fact that it can do it yeah, thanks to Gemini notebook LM for providing the code and the audio overview. You can play around with it on bindedkeys.com/pulseviz. There's a description at the bottom. Some of the maths a lot of the maths a future direction and yeah, you can check other relevant tools. Yeah, everything to do is available on bindedkeys.com. Go check it out. Provide your feedback and I'll see you next time. Bye.