Successfully Launching AI in Medicine

Carlos Ponce (00:00):

Good morning everyone. It’s Friday. It’s the last show of the week. And well, today we're going to be joined by James Hamet, the CEO of Vistin Labs. And of course, as ever, thank you Wade for, being here as a gracious. <inaudible> So I'm looking forward to this conversation, as a personal enthusiast of neurology and all things technology and the combination thereof. So, James, thank you for joining us today and welcome to the show.

James Hamet (00:43):

Thanks for having me, Carlos.

Carlos Ponce (00:46):

Absolutely. It's our pleasure. Alright, so we're going to start, with you, James. So let's start, start with you. Tell us a little bit about you a little bit about your background, where you're coming from, you know, all these things. So tell us a little bit about you, and then of course, tell us about Vistin Labs, and then we'll go from there. Thank you.

James Hamet (01:03):

Yeah, sure. So, I have a pretty crazy background because I'm an American. But I grew up in a French school. I went to French school until about the college, college level, and then in college, I went to Michigan. So I am very used to traveling, very used to being in unfamiliar environments. You know, my parents are both physicians. My uncle's an electrical engineer, and so I tried to marry those disciples. I, I've always been building technologies ever since I was a kid. I started making my first video game when I was 10, and I similarly became fascinated with neurology. I am a big fan of science fiction, you know, I, I read all the Isaac Asimov's and Robert Hyland. One thing that always, always got my attention was this concept of a hospital bed in a spaceship that you could lie down in. And it knows everything that's wrong with you, and you can get treatment help the same day, you know, the bed itself. Sometimes maybe you could even do the treatments. So I, I thought that that's a really cool technology that I want to make real. And I start, I started with the brain.

Carlos Ponce (02:16):

Absolutely. Well, that's quite an undertaking James. So anyway, you'll tell us all about it in a minute. So how about, telling us a little bit about Vistin Labs. What was the aha moment in which you felt, okay, we should this is going to start the journey of building the company. So can you tell us a little bit about Vista Labs and how it all started?

James Hamet (02:39):

Yeah, of course. So my dad's a neuroradiologist actually, so I already knew some of the challenges that face neurologists today. I also knew that neurology is currently using technology that is, you know, from the 18 hundreds, from the 19 hundreds, you know, spinal taps are over 200 years old. So it's pretty wild to me that the current state of the art is so dated. I thought, okay, there has to be an opportunity here to make things better. And, you know, my dad told me about how he's doing spinal taps on patients, and I learned that 40% of patients who even receive a spinal tap, don't get conclusive information back. So even though that's one of the gold standards for evaluating something like Alzheimer's Disease it still is not always informative. And of course, you know, you also have patients who suffer from the spinal tap. You know, it, 80% of them have headaches. There is, an argument that maybe for the headaches it's because the spinal tap was poorly performed. So, you know, someone who's ex experienced in spinal taps, maybe for them the rate is much lower possibly zero. But for patients today, you know, the rate of incidents of these unfortunate outcomes is still very high. And so I thought, well, there has to be a way to replace this 200-year-old technology. And that's when I came up with Vistin Labs. I had already been, like I said, developing startups. In my last company, Durable, we were making mind controls, technology, mind control, cars, mind control wheelchairs. You could know, our main product at Durable is controlling music with your mind through headphones. One of the things that, that I was doing there is I was looking at how people's brains respond to arrows. I would show an arrow on the screen and see if somebody, you know if somebody would have a special response in their head to a left arrow versus a right arrow. And how would that let me know, okay, they want to move in one direction or the other? I noticed that every single patient, every single person had a slightly different electrical response in their brain. And so that got me thinking that perhaps it could have something to do with underlying conditions. Perhaps somebody with you know, developing Parkinson's disease or Lewy body, they will have a different electrical response to shapes. And so I started investigating this pathway, became connected with scientists around the world you know, , to share my idea with them. And ultimately, I found a professor of psychiatry at Columbia who had started on this path. And, you know, we collaborated. We got a study finished. What we were able to discover is that yes, indeed you have certain parts of your brain that are responsible for interpreting certain parts of an image, just like, you know, a computer vision AI does today. You have different nodes responsible for edge detection et cetera. And so we found that, yeah, different parts of the brain respond to different types of imagery. And if the person's brain has been damaged in some way, well, their ability to respond to that imagery is reduced. So I created this company called Vistin Labs, and I'm excited to say that this technology is 30 days now away from being launched with our, with our first commercial clinics, where we help patients basically get their answers, their, their questions answered in their first office visit. So just like that hospital bed on the spaceship, imagine you can walk in to see your doctor, and instead of doing the common, you know, MMSE test, which is a hundred years old, where you're asked to count backward from a hundred, you're asked, you know, what year is it? Really ridiculous questions. In my opinion, everyone's going to get them right for at least the first couple of years. Instead, you come in to see your doctor and there's a brain scan device right there. You use it within 30 minutes of saying, Hey, I think there's a problem. And at the end, there's a patient report and it says, you know, these are the different things that you should check out.

Carlos Ponce (06:39):

I can only imagine the endless possibilities in unveiling the mysteries of what goes on in the brain and how that's going to help overcome, you know, many challenges in medicine and certain conditions that are, that people facing a lot of people out there. And before we get into the topic there's always just a little personal note, because I've always been fascinated, but by something that I read somewhere. It's called Dream Recording Technology. I mean, not anything medical about it, but it was, I saw something in there that the Japanese are experimenting with it. So I've always wondered if subjective experiences can be somehow interfaced with, with technology. Maybe we won't have the answer right now, but anyway, I'll keep an eye on that to see what happens. I'm fascinated with that possibility. Tell us a little about it, we're going to be talking about AI, so successfully launching AI in medicine specifically. So, we're going to be discussing AI in medicine, patient protection, and clinical excellence. This is the topic chosen by you. Tell us a little bit about why you chose this particular topic and why you felt it was relevant for today's day and age. Let's continue there, please. Thank you.

James Hamet (07:56):

Yes. Certainly, you know, the name of your podcast Tech Leaders Unplugged. You know, I get the impression and also speaking with you, of course, but you have this brand, you have this concept here for the podcast of talking about what it means to be a leader, and what are the things that we think about as leaders in our space. When I think of that, I think of, well, AI is a hot buzzword right now. People are probably interested to learn how we think about AI. AI could be a dangerous tool if it's not used correctly. AI could also be a useless tool if it's not used correctly. And so I picked this topic because I'd like to share my own perspective and opinion on this. You know, one of the things that I see, for example, as a common use of AI is figuring out if somebody might be sick or if they're not sick. I also see uses of AI to see if someone is at risk of getting a disease, and you know, how far away they are from getting that disease. And, you know, I think the first one is not very useful. The risk progression, I think is a little bit more useful. You know, if you can time when to do a checkup on a patient, I think that that is that that is much more useful. But ultimately, you know, who are we, who are we serving with AI? And I think that if we, if we want to help medicine, then we have to make tools for doctors. And when you have a tool, for example, that says, you know, this patient might be sick, that's not helpful to a doctor. You know, like the doctor, I've talked to doctors about my technology, you know, they're, they're not interested in how it says that this person might or might not be sick. They're interested in how we are predicting which tools to use with the patient to have the most success, to have less inconclusive results, to have, you know, greater patient outcomes, accelerated patient outcomes. So, you know, if, if you have a, a technology to my dad, for example, that says, hey, this patient might have Alzheimer's disease. If he agrees, then it doesn't add value, right? Because as a doctor, he already thinks that that's the case. The AI saying it's probably the case doesn't really add any certainty because it's not a clinical endpoint, right? And we rely on clinical endpoints to make diagnoses. If the AI disagrees with someone like my dad a, with a neurologist, well, what is the point of that? Because the doctor still should trust their gut. There's not necessarily a clear action to do when the AI disagrees with you, you know, who's, who's more right than the other. It's not like there's a debate that happens, you know? And, then you also run the risk of potentially introducing a malpractice situation for the doctor. What if the AI was right and the doctor was wrong? Right? That's, that would be very sad for the doctor. And so these are just some things to think about when we think about how AI serves medicine. I think that it's important to make sure that you're always serving the physician first. And in that context, you have to protect the patient's information. You have to make sure that you are improving, you know, clinical excellence. And the reason I put that in the title is because for me, clinical excellence, you know, it's the quintuple aim. It's, it's reducing costs, it's increasing health equity. It's making sure, that the outcomes are more actionable. That, I mean, my true goal really is to make it so that patients don't end up paralyzed. And, you know, all of these neurological diseases, get progressively worse and they lead to some form of paralysis where the person is basically unable to live by themselves. So, I would love to turn neurological problems into how we are currently treating cancer. For example, if you find the problem early, like a small tumor, you solve it early, and then the person can continue living their normal life. You don't have to wait seven years, which is, by the way, the average between when you first realize there's a problem and when you finally receive the right treatment.

Carlos Ponce (12:04):

There you go. Thank you so much, James. So, Wade, I know you have some questions, so please, by all means.

Wade Erickson (12:12):

So you know, the idea for this, you talked a little bit about where that came from primarily through your experiences with your father's you know, history, his work and career, and then some of your technologies. Tell me a little bit, bit about, you know, how your conversations with doctors as you were developing the features for this and some of the you know, product development application of techniques. Did you, as you started to build, I mean, obviously it's about ready to launch, so it's pretty well completed, validated, tested, you know, how much did you have to use, you know, does this require an FDA, you know, verification and validation? Tell me a little bit about that.

James Hamet (12:56):

Yeah, so our product is registered with the FDA, you know, we're fully FDA compliant. We're reimbursable as well. We technically haven't used any of our reimbursement codes yet. So, you know, when we launch in 30 days, I'll let you know how that goes. But yeah, you know, when I speak with physicians, the main responses that I get are that this is a tool that they wish had existed for, you know, since, since they first got started. Inconclusive findings are one of the biggest challenges in neurology today. It's extremely frustrating. And, you know, it's, it's also one of the points of contention between different types of physicians. You know, if you are, if you are in a, if you're a surgeon, if you're a cardiologist, you know, it's a lot easier for you to find out what's wrong with the patient. It's a lot easier for you to take action. And in neurology, you know, a lot of these people are more like detectives. They, study the patient for a very long time. They try to figure out, you know, what the person's problem is. And it's this long guessing game where they're wrong most of the time. And it's to no fault of the clinician, it's because they don't have the right tools. You know, the brain is very complex. We are only starting to learn about it now. You know, when I say learn about it, I mean, maybe we're at 1% of what we can know about it versus cardiology is perhaps 90%. I mean, look, we have a, we have a, a heart rate monitor. Show me an equivalent for the brain. It doesn't exist. We have a blood pressure monitor that doesn't, we don't have that type of thing for the brain. So these are the things that are important to think about. I was speaking with a neurologist only last week, and she started crying because she told me how she has some patients who very strongly believe that they have Alzheimer's disease. You know, they've been coming to her since their early forties, and they have some symptoms, some signs. But, you know, with the classic hundred-year-old MMSE tests and some similar ones, you know, you're asking the patient questions, you're showing them pictures and seeing if they remember the pictures. And these patients, you know, even though they know there's a problem, the test says that they're perfectly healthy. And you see this most particularly in people who, you know, they have an active lifestyle, maybe they're athletes. You see this in highly educated people, you know, maybe with advanced degrees a test that's asking them to count backward from a hundred. It's just not a test that you're going to fail. Even if you can't remember, you know, anything that happened 10 minutes ago, right? You might, you might start to forget the names of people in your life. You might forget where the grocery store is in your neighborhood and how to drive there. And you can still count. So this is very frustrating. It's a very frustrating situation. And, you know, this physician had recently sent a patient to get a spinal tap, and, you know, the results come back. And it's not, it's not negative, it's not positive, it's inconclusive. It's, we don't know. We don't know when we should do the test the next time. We don't know what this means. Is it going to change the next time we do it? We just don't know. And the sad thing is we have so many different types of tests, so many different types of essays that you can do. You really need to know what is the right one. And so that's why I think we're filling a blue ocean need. We are the first tool that is able to say, Hey, this test, out of all those tests, is the one that will show you the most about this patient.

Wade Erickson (16:30):

That's great. Obviously as you, you, you built this product and this company some of your previous experiences and products and work, I think in your autonomous, well, in your fatigue, mental fatigue analysis that you did that's on your bio in your LinkedIn. You know, I was wondering how some of that work led to these thoughts on, on how this might be actually analyzed, because you have already been looking at the brain's responses to being tired while you're driving and how that could feed with maybe autonomous driving. And we obviously, the cars already know about, you know, their surroundings and it's supposed to remove the person from the experience. But I think we still are a ways away from that. And, and I've heard about people falling asleep and the car is driving and they're supposed to still be around. How do you see some of that, maybe your previous work fed into this and kind of shared a little bit to the viewers on, you know, past work can drive new work, you know?

James Hamet (17:38):

Yeah. That's a very good question. You know, I can't say that I have a PhD in neuroscience because I don't. But what I can say is that I have pioneered and led innovative work that has led to new discoveries in the field of neuroscience. And those discoveries have been published in the top journals. I, it, it's fun for me, you know, I get to be creative, I get to basically be a sponge and learn from all the sources of knowledge that, that I follow in the world. You know the top neuroscientists, the top neurologists, I'm able to talk to them and sort of piece ideas together. And, so that part's really fun. Between my last company Neural and my current company, Vistin Labs, I explored a couple of different just like fun projects. You know, one of them was, I was thinking about a friend who had been in a car accident, and I thought, you know, maybe this is a problem to society that's worth investigating a little bit of course with a neuroscience perspective. And I saw that you know, fatigued driving is actually one of the most, most common reasons for an accident. The driver has not slept well enough or has had trouble sleeping or something of this nature. You know, it's really tough in our market right now where the economies are really tight. A lot of people are doing the gig economy. Some of them are really pushing themselves to the limit, of what they should be doing. For example, if you only have four hours of sleep versus a full night of eight hours of sleep, there's actually an interesting effect that happens to your brain where you miss approximately one second out of every seven seconds of information. So, imagine that you're driving and you're missing one, like one-seventh of every second. It doesn't sound like it adds up to a lot, but it does. And it's not, it's not actually one-seventh of every second. I mean, the way that we've seen it is, it's more like, one second is missing out of seven seconds. So for seven seconds, you know, six of them are normal, and then one second is just for some reason, your brain is not able to keep up with what you're seeing. So subconsciously you still see the environment, but consciously you're not aware that, you know, you're consciously uncoupled at that point, so you might not see something and then crash. So I, I thought, oh, wow, that's an interesting problem. And I created a baseball cap that you could wear, made it for truckers, and I also made it for airline pilots, and it measures your brain activity. It correlates with you know, how, how much rest we expected that you might have had before how likely you are to fall asleep. And when and the goal was to use that technology to prevent roadside accidents. I, you know, I, I can only commercialize so many products at once. So that was a little bit more of a just a fun activity that I, you know, turned into a big research project with Austria with the university out there and got it published. But I do hope that someone picks up that work and creates a nice solution that just helps people die less.

Wade Erickson (20:50):

Right. Great. One last question I had, we talked about patient protection and data, and we know AI requires a lot of training. And I would imagine understanding the feedback that you're getting from your current product that you're going to launch had to have a lot of you know, patient data. What was their current state? How was your, you know technology observing that brain activity to present that as a situation that needs to be looked at by the doctor? Tell me a little bit about, I mean, and obviously, that data needs to be held over time because things can change, and so old data collection, you know, the person's diagnosis might change over time, and you'd want to actually feed that into the situation with the data that was collected. Tell me about the lifespan of, you know, we worry about a patient's privacy of their address that's external to them. I mean, this is their brain activity, and that's close to DNA almost. Tell me a little bit about some of those things you think about as you're training these products and keeping this data on for long periods of time so that you can learn from the past, and how the future might even change that, what you understood about that data.

James Hamet (22:08):

Yeah, that's very important. It's important to know that when the patient uses our products, that's not the only time point. That's interesting, right? We also want to know when they got there, their diagnosis. If they get one, maybe they had one in the past, maybe they get one in the future, you know, that type of thing. We want to know the results of blood tests, pet scans, MRIs. This is all information that helps us improve our product. So, we want to know if it happened in the past if it happened in the future. We expect that most patients who use our product, you know, like I said, it's physicians to be the first tool that's used in the clinic. So we expect them to come in without much data to contribute besides, you know, their, their brain activity that we collect and predict on. But we would like to keep up with that person. So we've made some predictions about them, right? We anticipate how they'll score in certain tests when they do get those tests, we would like to know. So one of the things that, that we have to be very that we have to be very confident about is how we are protecting the data, not just from, you know, cyber criminals and you know, to protect the protected nature of the data, but also to make sure that, like you said, that it survives that it's data we can use in the future. One of the things that we will be doing, of course, is timestamping every single endpoint knowing when it was collected from the patients. That's the time point that's the most interesting to us. The day, for example, of the PET scan itself the day of the, the diagnosis, the day of, you know, the MRI, we would like to know all the, all of that information. And I think that's how we'll make this technology become, we'll make our, our data that powers the technology become evergreen in a sense, as we continue to collect data continues to grow as long as we have the time points, then we know how the patient has evolved over time that allows us to yeah, keep the data relevant.

Carlos Ponce (24:12):

James, we're coming up on time. And I want to end it by, well, I have one more question for you. And again, as I said at the beginning, think of this as a question coming from, from the layman's perspective from the, someone who's really new to the technology. So I'm just thinking, I'm, what I'm hearing is that, you know, the, the, it's very easy for me to understand the tremendous possibilities, solving specific problems for patients, right? You know, unique way to say the least. But at the same time, we are talking about data that is being generated for medical purposes and all that. So when I hear the word data, the next big question, and this ties into your, what you mentioned about you're a big fan of, fan of science fiction. So in the future, like future scenarios, what sort of exchange or role can your technology, your solution have on big data, you know human behavior, anything about human behavior, analyzing patterns, trends, what sort of value, or if at all, if applicable, or what sort of future do you see in, in your solution in the, in the, in the realm of big data, if applicable?

James Hamet (25:31):

Yeah, that's a good question. You know, I'd say that I, I'd say that the way you make this is now leaning on, you know, my personal perspective on ai. But it's, it's been, it's, it's led me to much success. So, you know, the way you make AI useful is by having, of course, the best features. How do you get the best features? You know, you could take a whole bunch of noisy data and throw it at a neural net and then have it, you know, find latent variables. That's definitely one strategy. But I prefer the strategy of, you know, designing your system, your observations in a way where you're collecting, the richest features. And so for us, big data is not totally applicable because we design new paradigms. We design new endpoints that previously haven't and can't be collected. You know, one of the ways that my technology works, and one of the ways that makes us so different from other companies in the market that are also, you know, for example, looking at EEG is that we actually are stimulating the brain, and we're not stimulating the brain with an electric rod, or not even with magnetic stimulation. We're going into the eye. We're showing very specific animations. We're showing things, that we expect the brain will have to process in specific ways. And it allows us to characterize brain activity very, very well in a healthy person. And in somebody who's not healthy, we can see what differentiates two people, even within the same group of healthy or not-healthy individuals. So that's what I think is the most important thing here. Of course, big data will become a fingerprint for almost anything that you want to learn. However, at the same time, it's also fitting everybody into a square peg. And I don't think that that is when we're talking about precision medicine, I don't think that that applies here. You know, everyone's case is very different. You have very, very rare diseases. You know, some of them affect only tens of thousands of people. I don't think big data is the answer to those types of problems.

Carlos Ponce (27:51):

I see. Well I mean, we could think of a million more questions, but we don't have time, unfortunately. The only thing left for me, James, is well, thank you for having been with us here on Tech Leaders Unplugged, and I want to remind the audience that well next week we have several more guests. They're all going to be right there on the show's website. We're still configuring the setup of the, you know, promos and all that. So that's why I'm not going to display them here as, as it is customary, 'cause it's still a work in progress. But join us right here on Tech Leaders Unplugged. Remember, it's TechLeadersunplugged.com. So every weekday, almost every weekday at nine:30, just keep an eye on our website and it'll all be there on the show's upcoming calendar of events. That being said, again, thank you James, and thank you, Wade. Sure. And we'll see you next time right here on Tech Leaders Unplugged.

James Hamet (28:54):

Thank you, Carlos.

Carlos Ponce (28:55):

Have a good weekend, everyone. Thank you.