 My name is Dr. Surya Sanyal and I am the CEO of ThinkBioSolution and today I will show you some use cases of how artificial intelligence-powered medical devices are changing traditional healthcare pathways. And then at the end of the talk, I will let you decide if such technologies would partially or wholly change what we call healthcare today. I'm going to replace doctors and sports coaches. Are you okay with that? Good. So our story began, or my personal journey with healthcare began about 10 years ago when I was visiting my grandparents in India and my grandmother, she loves to cook and she was cooking for me. She slipped and she fell down. I took her to the hospital and I was shocked to see that the doctors and the nurses were super enthusiastic to help the patients, but they were overwhelmed at how much the patients demanded in terms of insight and care from them. So I walked back home that night thinking, would it be nice if the medical devices that my grandmother was tied to in the hospital could, in addition to monitoring her health, could tell her what to do. For example, when she should take her medicine or could call the nurses if her heart rate went up. The second thing that I was wondering was what if the medical devices could continue to monitor her once she comes back home and help her to do her exercises regularly so that she can get back well soon. So for the last five years, my team and I are building technologies in terms of software, hardware and data analytics and artificial intelligence to solve exactly these problems. Just on a side note, my grandma came back in a couple of days back from the hospital and continues to become to be the best cook in the world. So Thing by Solution is an original device manufacturer building privately labeled wearable medical devices and customized software for telehealth and professional athlete monitoring companies in US and Europe. The chief innovation officers of these healthcare companies love us because we allow them to add AI-powered wearable devices to their portfolio in the shortest amount of time. And you can see our little mannequin Tom here wearing one of the wearable devices. So you wear it using a strap, like the one that Stam's wearing. And it can measure biometric parameters and then sends it via a Bluetooth-based hub or a device to a HIPAA compliant and a EMR compliant cloud. The data can then be viewed by the user, the doctor or the sports coach using a smart dashboard on the phone or on a desktop. Our device has an embedded real-time operating system that has artificial intelligence capabilities that combine the biometric with the movement data to give real-time feedback. For example, how fast you should run to build endurance or about how fast you should run to burn your fat or it can give you feedback when you're stressed out or it can tell you how to do an exercise test test or it can tell you how much to sleep. But before I go into explaining the cases, I'd like you to remember one of the take-home messages. For example, our AI engine can help you build cardiac endurance twice as fast as compared to running with an actual coach. So this brings me to my first of the three case studies that I'm going to discuss today. And about five years ago when I came to Dublin, I saw that there is the Dublin Marathon and I wanted to run in the Dublin Marathon and even perhaps win it. However, as a grad student, I didn't have the money to afford a professional coach to help me with this. However, I knew that if I could run at my optimal heart rate and blood oxygen saturation levels at the optimum speed, I'd be able to build endurance fast at the desired level. And of course, I was also talking to my grandmother back then and she heard back from her doctor and the doctor said that the faster, if you could walk the maximum amount of numbers, if you could take the maximum number of steps per day, that will really help you to have a better health. And she was working with her physiotherapist to achieve her walking goals per day. So the point being is either if you are an endurance runner or a patient, you go back to your sports coaches or your physicians who in a clinical setting can help you achieve those goals by combining the health data with the movement data. And however, then, because in our device, we measure both biometrics and movement data in a real world, in an ideal world scenario, it was kind of trivial that we would be able to do exactly what they are able to achieve. However, when we started doing the project, we realized that in an ideal world, we'd like to collect lots of statistical significant data on each user every time they go out to train. However, users in a real world wanted to only train for about 15 minutes and no more, train the system for 15 minutes and no more. How do you bridge the gap between collecting statistical significant amount of data and the fact that users only want to train for 15 minutes? Well, the answer to that was artificial intelligence. We took about 10 participants in a program and made them run both for a long time as well for a short time that gave us the statistical significant data as well as the sparse data. We fed all this information in a machine learning algorithm that was semi-supervised and that spit out a predictor that could predict your optimal running speed based out of sparse data. And we believe that the proof of pudding is in actually tasting the pudding. So we went ahead and tested this with real users. So we had 10 users who for the first couple of months ran based on the feedback that their sports coaches gave them. And for the last month ran based on the feedback that they got from the wearable device. And at the end of each month, we actually measured their endurance on a treadmill. And it turned out that for every user has actually built endurance twice as much at the end of the third month when they were training with the device as compared to running with the sports coach. So that was the first use case and we believe that that was really significant progress into what these type of medical devices can deliver. But the second case study that I'm going to refer to you today is about stress. In a start-up scenario, we're always stressed about delivering product in time, delivering good quality of service to our customers. And stress is something that's not new to us. And neurologists have known about this for ages and they use multiple lead ECGs in clinical settings to measure stress by looking into raw heart rate type of data. However, what clinicians don't know is how to combine heart rate and heart rate variability information. What is the ratio of these two things that they need to measure to measure stress? And since we measured both of these things using our wearable device, what we did was we took 20 users and we placed them in a dark room and we showed them a black and white movie and then randomly induced stress of three kinds. One was a pinprick, two was certain stopping of the movie and three was introduction of a loud noise. And we also continuously measured the biometrics. And all of this information was then put into a deep neural network which then spit out a predictor that could measure stress by looking into heart rate and heart rate variability as an input. And this is how the predictor algorithm looked like, if you will. And then we went ahead and retested this predictor with 10 new users using a different movie and slightly changing the stress conditions like what was the intensity of the loud noise and as such. And however, this time they were wearing the wearable device in real time and every time the device thought they were stressed out, it buzzed or it gave a haptic feedback. And we measured what was the positive, what was the percentage of true positives or what is, I mean, what are the number of times as a percentage that people responded to true stress. And we measured this for our predictor. That's what you see in the red line. It was about on average about 84% accurate. And we also measured what happened when you just used heart rate variability which is the industry standard to measure similar stress with the predictor built with just heart rate variability. And that was about 41% on an average for the 10 users. And that's the blue line. So take home messages, machine learning can help us build predictors that measure stress better. So the third case study is also inspired by the story that I told you about my grandmother that she slipped down and fell. And oftentimes what happens is when geriatric patients fall down, the ambulatory care person don't know how to respond, what care path they do take for them. And let's say the person had a stroke and the ambulatory care person thought that he slipped and fell because of slippage and starts not giving him the care that he or she requires. And this can actually lead to loss of crucial time and even leads to death of patients. And of course, because our device actually have a built-in motion sensor, a nine degrees of freedom motion sensor, we thought that we could actually solve this problem. So what we did is we made people wear this. We had 15 volunteers who wore this device and either sat down or fell. And we fed all of these data to a multi-scale clustering algorithm and allowed the mission to build a predictor to differentiate between a sit-down and a fall. And so this is a project that's still running. And then we used the predictor on five users to see what was the false positive, how good was the predictor. And as you can see, in about 82% of the cases, the machine could actually differentiate between a fall and a sit-down. At the moment, we are adding biometric signal on top of this data to help the machine differentiate between a fall and a cardiac event. And that would be what would be a really value add to the current neurological fall industry out there. So this kind of brings me to the end of the talk. And I have shown you three use cases that we believe are just instances of how wearable medical devices powered by AI can change care pathways. I definitely believe that combining these sort of technologies, along with continuous monitoring, will help people automatize primary care. And I would like to hear more about how do you think this will affect health care in your own ways after the talk. Please feel free to come to me and talk about that. Thank you.