Podcast Transcript: Empowering Doctors with Artificial Intelligence

By Sacha Heppell - November 04, 2020

 

Empowering Doctors With AI - Interview with George Hauser and Artur Adib

PODCAST NOTES

In today’s episode we are joined by Artur Adib and R. George Hauser of Biocogniv. They discuss the advancement of AI, how it will benefit physicians and address unmet needs. Artur begins by discussing his background in AI and explaining what inspired him to pursue Biocogniv as founder CEO followed by George who shares his background and his story in machine learning and how the platform at Biocogniv works. Artur explains what the platform provides for physicians. A marriage of software engineering and academic research. He goes on to talk about about scaling up their technology by making sure that they are looking at as many hospital systems as they can and doing deep analysis of how the algorithm performs in different population groups by demographics. Artur explains that a key issue in advancing their build out would be understanding that if you are connecting to a hospital system, making sure this is a secure connection that will protect patient data as well as prevent hackers from entering that hospital system through their system. The deployment strategy has to be safety and security first.

George begins sharing,  “Anytime you start to have those kinds of input from different sources of information you start to get a certain amount of complexity, AI is good for sorting through that level of complexity and integrating that to optimize the patient outcome.” He sees virtual care as speeding up the momentum for AI as it has become a true delivery model. He goes on to talk about how one area that will help empower doctors is the integration of computers with healthcare providers versus the replacement of healthcare workers. AI does somethings well but may not be able to solicit information from someone who is reticent to give that information but a computer is much better integrating a large amount of information. They have to work together.

PODCAST TRANSCRIPT


Sacha Heppell

Welcome to Who would have thought my name is Sacha Heppell, Chief Marketing Officer of SmartTab. And I'm hosting this podcast with Robert Niichel our Founder and CEO. Robert's experience and leadership and management of pharmaceutical research and development led to the founding of SmartTab in 2016 to combine wireless technology with pharmaceutical drug delivery. Today we explore artificial intelligence, otherwise known as AI, and how it can empower doctors. We will speak with an extraordinary machine learning specialist and AI mastermind about the current and future applications of artificial intelligence for COVID-19 and beyond. I'll pass it over to you, Robert to introduce our guests today, behind the artificial intelligence revolution, Artur Adib and Dr. George Hauser.

Robert Niichel

Thank you, Sacha, I'd like to introduce Artur Adib, PhD Founder and CEO of Biocogniv Artur has built the flight controller software for the world's largest electric flying car and was a senior engineer at Twitter, a former faculty member at the National Institutes of Health and leads the vision and AI for Biocogniv. We'd also like to welcome Dr. George Hauser, Chief Medical Scientist at Biocogniv. Dr. Hauser is the author of over 20 peer reviewed publications, and specializes in research involving artificial intelligence and electronic health records. At Biocogniv, their model for COVID-19 screening is the largest application of artificial intelligence to COVID yet, and they're leading the push to empower doctors with AI. Okay, good morning. Hello, Artur and George, welcome to the show. Thank you for joining us today. Artur, can you tell us your story, your background in AI? And what inspired you to pursue Biocogniv as the Founder/CEO?

Artur Adib, PhD

Yeah, absolutely. It's a great pleasure to be here, Robert, and Sacha, thank you for the opportunity. My background is, I came to this country for graduate school in computational and statistical physics about 20 years ago, that's back when AI was sort of a fringe topic was mostly talked about in academia, to my knowledge, at the time, there was very little being done outside of academia. Biocogniv started out of my catching up with the literature of the last few years of the medical literature, applying AI into diagnostics in particular. I have even framed here in our office, some of the early papers that that I read that really left an impression. Those were papers that applied a state of the art machine learning to image classification in the specific case of pathology. And I found that it was one of the most powerful things that one could, simply by presenting examples of disease state images, and healthy in that specific case for skin conditions, skin lesions, that you could train a model simply by showing examples and it outperformed, in some cases, dermatologists. And I felt like that was such a milestone for AI 20 years later, that I wanted to be part of it. And as we now know, there's quite the conversation about AI in this space. And so that's sort of the the gist of how that all started.

Robert Niichel

And then moving over to George, could you share with us about your background and your story in machine learning and and how the platform at about Biocogniv, how that works?

Dr. George Hauser

Yeah, I echo what Arthur said, I'm happy to be here. Thanks for putting this together. I think it's great to have people start discussions around this and see where it leads us. So my background is in biomedical engineering. And when I started undergrad, wasn't that long ago, and there was only a few biomedical engineering programs in the country. And I went to one at the University of Michigan. And when I graduated, I went into medicine. And I was using a lot of just like computer technology in general. And it felt like there was a real need to be able to synthesize all this information that was being and I still think it is, poorly organized and just just saw an opportunity there. So graduated medical school, and then I went to train as a clinical pathologist, which is subdomain in medicine that focuses on laboratory testing, and in that space, I've continued my interest in AI. And then so when when Artur came up with his proposal, it just seemed like a natural fit. So you know, we're working together in this space now.

Robert Niichel

And then two questions. The first one is, what exactly does your platform around the AI? What exactly does it provide for the physicians?

Artur Adib, PhD

Yeah, so the company started last year. And one of the things I guess I didn't mention is, after graduate school, I spent several years at the National Institutes of Health as a faculty there, and then spent some time in Silicon Valley, I decided that academia wasn't quite right for me, and then spent some time Silicon Valley at a company, companies like Twitter, Magic Leap, and more recently, an aerospace company. But the way I think about Biocogniv, is this marriage of software engineering, and academic research, and when I started working at Biocogniv last year, obviously, there was no Coronavirus, and the goal of the company was to identify unmet needs, in hospital systems in particular, tertiary care, and try to help address those unmet needs with AI. So we had some targets we started working on before Coronavirus hit, but then it happened, COVID happened. And we saw how, early, how much of an impact it was going to have in hospital systems here and whatnot. So I think this is a testament to the power of AI, combined with the availability of electronic medical records. So we pivoted our efforts towards COVID. And we were able to come up in probably less than five months, we were able to go from zero to a model, product, that could rule out COVID with extremely high sensitivity based only on blood test data like complete blood count and comprehensive metabolic panel, which are routine labs that are performed at hospital systems. Again, I think this is a testament to the power of AI. Typically when when one wants to come up with a new biomarker, traditionally, this is a multi year process tens, maybe hundreds of millions of dollars to understand the pathophysiology of the disease and then identify the molecules that take part in the process and then isolate the molecule and then come up with an assay for it and so on. I think this is a great example of how AI can be really helpful, like George was alluding to, there's hospitals in the country, in general, has a ton of data already. And it's generally poorly organized and poorly leveraged. So we think that AI can really unlock the power of that data and in what we call AI-COVID. That is our first product and is a great example of how that can happen. And answering your question around the utility and how it can be used in clinical practice. The main reason we did this was in response to the ongoing shortage of kits, of testing kits, across the country. So FDA has approved over 200 test kits. Despite that there is an ongoing issue with supply chains across the country, it varies from week to week, one week, a hospital system will have plenty of swabs, plenty of reagents, and so on. But another week, they will run out of one of those components. And that will set them back. Obviously, AI doesn't have that issue in particular, especially because AI in particular lives on top of very, very routine laboratory data that you generally don't run out of reagents for so that's essentially it. That's the use case is tertiary care hospital systems, is to help them triage really COVID patients when they're struggling with their PCR kit supplies.

Robert Niichel

Talking some more about the COVID-19. Like you said, you went from zero to you have on your web page listed over 60 US based hospitals, over 200,000 patients. That's a significant, that's pretty heavy lift. And so, maybe you could talk about some, number one, challenges through that. And then, it looks like COVID-19 is going to be around for a while so how do you see your technology being continued to be scaled up? What are your plans there to, you know, say move it from 200,000 patients to a million to two million etc?

Artur Adib, PhD

Yeah, great question. I mean, obviously, there's a lot of papers, a lot of AI work being done. To your point, like we think that one of the key things that sets us apart from other studies is the sheer size. The multicenter nature of it, 66 hospitals. Some of it was heavy lifting in terms of getting IRB approvals with hospital systems and whatnot, but the vast majority was through a vendor or a partner and we believe that it's the largest, perhaps not only for COVID, but probably for infectious diseases application of AI to infectious diseases and whatnot. In terms of multicenter, we can spend some time talking about the, you know, as you probably know, there's a lot of concern around the blackbox nature of AI, especially physicians and clinicians having the ability to understand what is going on, before they accept a certain result. A great, probably the largest part of earning that trust with the community, the medical community is making sure that the product generalizes well to different populations in different health systems, often, you will see models and AI models that have been trained on one or two hospital systems that are not for a variety of reasons, not representative of the wild, right. So the the field in general, where it would be applied. And so we believe that we can mitigate some of these concerns by making sure that A we look at as many hospital systems as we can but also B do a deeper analysis of how does the algorithm perform in different population groups by demographics, for example, is one one way that we look at it?

Robert Niichel

And then let's talk a little bit about, how you expand. So you started in a certain way, and then COVID-19 came along, and you're focusing on that? And then do you see this technology as you expand it being more demographic focused or on particular, right, so COVID-19 is a particular disease or virus, you focus on that? And then is there like something else, like and does this get broken out per disease? Or how do you see it expanding per category?

Artur Adib, PhD

Like you mentioned, we think that COVID is gonna be here for a while, unfortunately. So that's keeping us plenty busy. But that said, we started the company with a with a bigger mission. And we continue to believe in that mission. And that is that there is a variety of conditions for which AI can deliver value. I'll give you one example that we're particularly interested in, that will be pulmonary embolism. If you take something like Troponin, that is the biomarker of choice, obviously, for myocardial infarction, virtually every patient that presents with chest pain in the hospital system will get your Troponin ordered. We try to be the Troponin for pulmonary embolism. Yes, there's a marker for ruling out PE, D-dimer. But it doesn't have the same characteristic performance as Troponin. So it's not as widely used, we believe that we can move the needle on that. And as a result of that, we can avoid the over utilization of CT scans, for example, which is considered by probably the majority of emergency physicians, one of the top issues that they're facing is again, the over utilization of imaging, which has been actually reflected in how payers see this. The largest payers in the country are not paying for outpatient imaging, CT imaging and it's not getting reimbursed because they feel like it's been over utilized. So we think we can help with that.

Robert Niichel

And then as you expand the pulmonary market, that's a huge market right there, just that one. And then do you go back to these same 60-66 hospitals you're in? Or do you focus more on ones that are more advanced with cardio or emergency rooms? Or how did how does that roll out?

Artur Adib, PhD

That's a great question, I think you're getting to sort of the core of a successful AI product. And that is, it has to be representative of the data that it will actually see in clinical practice. And one way to do that, obviously, is to just try to get as many hospital systems as you can, not only in different geographic regions, but also in different resource sort of circumstances, like small community, rural hospitals, versus large academic centers, and so on, because they, they'll have different instruments, they will have different practices and whatnot. So you want the model to be robust against that. We are working with some collaborators at different institutions, this will probably provide them the most depth in terms of the quality of the data. But it's impractical for a startup like ourselves to go after, let's say, 60, hospital systems one at a time, getting one IRB at a time, and so on. And so we are lineal partners, and so on for this.

Robert Niichel

The larger strategic partners are always helpful, especially if we get the correct one. So I have one last question here, and then I'll turn it over to Sacha. So really, where do you see your technology as it rolls out, how it gets built out? So number one, your technology, specifically in the next five years, and then in general, more just the entire AI market? What does that look like in five years from now?

Artur Adib, PhD

Yeah, that's that's a timely question. How do you roll this out? So I was just reading the news yesterday and today. The country has been hit with a massive cyber attack in hospital systems in the country. Over 400 are being hit with cyberattack, including one here in our backyard and the University of Vermont. And we are very, very aware of this, we are going to be a software based deployment in hospital systems. And we have hired two of some of the the top engineering talent out of Silicon Valley, including one who's an expert in security. So I would say that one of the key issues, it's not talked about too much, in addition to all the blackbox nature of AI and all of this. One of the key issues will be if you're connecting to a hospital system, how will you make sure that this is a secure connection that not only will protect patient data, but it'll prevent hackers from entering the hospital system through your system. And so we're very much on top of this, it's one of the top concerns that FDA has. So we've been having almost bi weekly meetings with FDA to get our product through the finish line. And this is top of mind for them. And right now, like literally in the last 24 hours this this massive breach has happened, I think it's gonna only going to become more important that we address this. So I would say that answers a part of your question like our deployment strategy is safety and security first, the second part is projecting out into the future. I'm extremely optimistic about what's what's happening with AI across the board. I know that there's there's some well received skepticism. And I think that's the that's necessary, it's healthy. It forces us to think hard, long and hard about how we can prove safety and effectiveness, which is sort of FDA per view for medical devices. But 5 to 10 years from now, I think it's easy to imagine a world where radiology for example, right now, you don't even have to project that out that far. Like we're already starting to see unprecedented reimbursement pathways, a company like Viz AI, for example, in the last month or two has received reimbursement as sort of a reimbursement pathway from CMS that for their pure AI software, which I think is unprecedented. I think we're going to start seeing a lot of folks are saying that this has opened the floodgates for AI reimbursement, I think we're going to start seeing more and more of this, that's certainly a path that we want to pursue, we want to make sure, ultimately what we feel is that AI has the potential to become as safe as effective as any other medical device, we just have to show that that's the case. And I think that there's mounting evidence now that that's becoming the case. And so in 5 years from now, 10 years from now, I think pathology is going to be, and I'll let George speak somewhat to this, as a pathologist, he knows that there's a lot of activity in this space, radiology is another one. And we think Laboratory Medicine, this is sort of our niche, folks talk about 70% of medical decisions being made through laboratory results. And imagine how much you can, how much farther you can go by combining multiple laboratory results, and analyzing them and identifying patterns that humanly were nearly impossible to identify these patterns. And so that's essentially what our product for COVID has been able to do, is to identify patterns in routine laboratory tests that doctors are used to seeing every day. But by presenting large numbers of these results, and ground truths, as we call them for who's sick, who's not, then the models have been able to pick up these patterns. So I think we're going to start seeing more and more of this, and how that helps hospital systems in a variety of ways. We already talked about avoidance of CT scans and whatnot, we think there's also an opportunity to to expedite results. Cultures, for example, are notoriously slow, they take on the order of days, bacterial cultures and whatnot to come back. Potentially, we could shorten that time to minutes or hours and really change patient outcomes there. Because with that information, doctors can intervene accordingly.

Robert Niichel

Well, that's, that's great. Now, that's it for my questions. And then Sacha, go ahead.

Sacha Heppell

Thanks, Bob. So George, I'd like to ask you about what do we have yet to discover? You know, as you're a machine learning specialist, I just like to know, kind of like what our Artur was just saying about identifying those patterns and all of that data coming together. What do we have yet to discover about the benefits of that, of this technology?

Dr. George Hauser

Yeah, I think you can take the technology in a lot of different directions. I think one of the avenues that we're taking it in is sort of this ability to find patterns and which otherwise may be interpreted as noise, in the diagnostic realm. But I think in general, you know, depending on how you want to box in AI, it's going to be used in a lot of different ways, throughout healthcare, in the diagnostic realm, which we've been talking about now, but in a lot of other ways, including more mundane things like scheduling, workflow, and I think you're going to start seeing it more integrated outside of this sort of Crystal Palace and more democratized to things like visits online, no sort of virtual doctors. And then that's a huge shift for medicine, which has traditionally been a very sort of brick and mortar type of experience that that people have had. And I think there's a lot of change that's going to happen, you know, it's sort of been rippling through a lot of different fields, you know, from finance for quite a while and moving into medicine more recently. But I think I think we're just standing on the tip of the iceberg here, it's going to affect essentially every process within a healthcare system, at some point.

Sacha Heppell

Yeah. And focusing on one of those areas that actually we work on here at SmartTab, we're developing drug delivery systems that would work inside of some of these AI systems. How do you see that kind of playing out? How would this potentially integrate with digital medicine?

Dr. George Hauser

I think the system you're describing has a lot of different sources of information from like image analysis to patient records, diagnoses. And anytime you start to have those sort of inputs, you get a certain amount of complexity. And that's what AI is good for, is sorting through that level of complexity. So you know, one of the additional streams that you could add to that would be some of the diagnostic products that we're developing and integrating that with, with what you're doing to optimize the patient outcome. That's, I think that's what we're trying to do here is about that, that data integration step and how to do that efficiently. When it becomes overwhelming. I mean, you guys are probably taking quite a few images, as your device travels through the body and being able to process those, analyze them and optimize it is a difficult task for a person to do by themselves.

Sacha Heppell
And so then looking at, you know, virtual care as it's become a true delivery model. Now, how do you see AI really supporting that in the short term, but also down the road? And like, how this will evolve?

Dr. George Hauser

Yeah, sure, I think it's probably going to speed up the whole transition to AI, you know, if you want to think of AI is like a train, you know, it has a certain amount of momentum. And I think early on the gatekeepers of that momentum, we're not familiar with the technology, it was sort of foreign to them. And so it kept the pace of the train moving pretty slow. And as the the new leadership comes up, and they've grown up with these devices and and use them, you know, basically their whole lives, I think the speed of the train is starting to pick up. And as we transition more to virtual care, what that's doing is taking away the elements that may have been traditionally difficult to capture in sort of structured data, things like the physical exam, for instance. And then that dialogue that you have with the patient. So what the virtual care is likely doing, if you're looking at the big picture of AI, it's creating more data. And as you create more data, and especially in a in a way that can be analyzed, then that's going to be able to feed back and train the algorithms and allow them to be improved. And just keep the train moving. And I think it's starting to pick up the momentum from that. And it'll likely keep doing so for for a number of factors, from just keeping costs in line to outperforming current standards of care. You know, healthcare is a very human intensive field. We have a lot of support staff, nurses, technologists, technicians who have specialty training and in each of those fields, and you know, AI certainly has a lot of different niches that it can fill in any number of these applications.

Sacha Heppell

What is one that you see is like the most important that would really that's really going to empower the doctors and in the work that they're doing. Which area would you say is like the most important?

Dr. George Hauser

I've always tried to try to stress the integration of computers with with healthcare providers rather than use words that suggest a replacement. Because I feel like AI does dumb things well, and people do some things well, so a computer might not be able to solicit information from a person who is a little reticent to give that information. But like I said, computers much better integrating the large amount of information. So I think they have to work together. And I think you've seen some instances where, like with the Watson, I think one of the ways that they sort of approached it the wrong way was they try to pitch it as sort of like a replacement for the for the doctor. And I think it didn't, it didn't work out so well in that sort of pitch. And remember watching the Watson performing the Jeopardy against Ken Jennings, I think you can still find that on YouTube. And the clip that stands out to me was this point in time where, the question was asked, and then somebody rang in and they gave an answer. And then it was wrong. And Watson rang in right after, and gave the wrong answer, the same wrong answer that the person before it had done. And so it seemed like they hadn't trained Watson to listen to what the other person was saying it was more sort of like a no give its answer independently. So you know, the AI can fail spectacularly. And I think it's important to keep in mind that we're all working together, and everyone has sort of their advantages, and just sort of go about it with that general mindset.

Sacha Heppell

Yeah. And as you develop your system, are you how do you keep that mindset alive? While you're innovating your system? Like, what what are some ways that you make sure that you're really having it be physician-centered, or patient-centered, that it's going to actually really make a difference, and not just be a great piece of technology?

Dr. George Hauser

Yeah, that's the goal. That's the holy grail there is to integrate it in with what you with, with the healthcare environment. You don't want it to be this sort of like warp that people don't want to touch. It needs to be integrated well, and you do that with getting feedback from the people who are going to use it, and making sure that they believe in it. And when they give you honest feedback, that it doesn't work, you try to improve it. I think one of the things that we really strive to do is to actually change outcomes. And it's very hard to change clinical outcomes, whether it be extending life, or whether it be improving quality of life, or whether it be, you know, showing a clear monetary savings. Those are all very hard outcomes to achieve. And I think working towards, you know, proving that an AI system can do that, in a real world setting. That's what needs to be demonstrated and demonstrated repeatedly and in many different areas, to keep that train moving as fast as it can.

Artur Adib, PhD

Yeah, I would echo that, I think that there's a lot of, there has been an attempt to bring AI and I think the Watson might be an example of that, where it's, and maybe Google, you know, it's a catch all type of thing, like you're bringing AI, and that's going to be this massively disruptive thing that is gonna, you know, do XYZ all at once. And so one, we feel like, it's, we're gonna have to win this one battle at a time, like, like George was saying, like, boy, it's so hard to prove that he actually moved the needle on outcomes. So I like to think of us as more of a diagnostic company first, that happens to be using AI. And so that keeps us more grounded in terms of the deliverable to our customer. It's not just a fancy gadget, it's something that will actually improve the quality of your diagnosis. But you can't do that. It's incredibly hard to do this. If you try to do everything at once you got to you got to carve out every little unmet need at a time. And boy, does it take effort to prove that you can actually do something in that little space. Right. But I think that ultimately, that's that's going to be how the the puzzle is going to be solved. It's like one piece at a time and not like everything at once, if that makes sense.

Sacha Heppell

Yeah, totally, totally makes sense. If you put yourself in the shoes of a doctor, maybe there's an example that you could paint a picture? What would that look like, as they're doing their work?

Artur Adib, PhD

I can speak to for example, the the COVID product that we have today. It's going to be seamlessly integrated with the Electronic Health Record system and we think that all of our products are going to be like that. So it's not going to be the case where a physician is going to have to interrupt their workflow and to go learn like an app or whatnot. And they're going to be able to place an order just like they place an order for laboratory tests today. And rather than that taking however long that would normally take, you know, we're seeing some some hospital systems, taking days to get a result back because they have to do a send out and so on. They're going to be able to just do a routine blood test and get the result in under one hour. So from their from their perspective. If it will work just like any other lab, except that it will be a faster turnaround, and potentially even more accurate than, than the existing available tests.

Sacha Heppell

And so what's next for Biocogniv? What do you sees the next steps for you and what your focus is, like you say, focusing one step at a time, what's the next steps for you?

Artur Adib, PhD

Yeah, we're laser focused on getting this product through the FDA. We want to really help with the current pandemic, like we talked about, there's, this is not going to go away anytime soon. And so we want to make sure that we can help and move the needle on testing capabilities across the country. And the first step towards that is really to just prove the safety and effectiveness of the product with FDA. Then obviously, pursuing reimbursement strategies and so on. But then we do have a pipeline, and we like to think of us as more of like a machinery that produces these models, reproducible machinery. And so we're building a system internally that will enable us to more efficiently produce these models, whether they're, you know, for infectious diseases, or for cardiovascular conditions, like we talked about shouldn't matter, but we want to build a platform that can actually produce them. And so we have sort of the alpha version of that, that we've used to, to build AI-COVID. But ultimately, we're building this bigger platform for for other conditions. But like I said, I think it's really important to stay focused, one piece of the puzzle, at a time. I think COVID is gonna keep like 90% of the company busy for the next year or so. And then the remaining 10% is going to be like one step forward thinking about this platform and in the new indications.

Sacha Heppell

Thank you for taking that on and your commitment to really focusing on COVID-19. And, and making a difference for transforming healthcare and your commitment to advancing this technology for that. And we really, we admire that and and your team for the work you do in innovation and sticking to it and figuring it out. You know, this is so we really appreciate that. Is there anything else that you'd like to share with us?

Artur Adib, PhD

No, I would just echo what you're saying, we just went through Y Combinator over the summer, which is probably the number one startup incubator in the country, and I got to meet phenomenal entrepreneurs there. I would just say that, it goes right back to you guys. Innovation is so incredibly hard. And I have great admiration for people who go out of their way to pursue innovation under very uncertain circumstances and so on. And so it's a rough road, but we think it's a it's a worthy road. And kudos to you guys. I mean, the the product you guys are envisioning sounds out of like a science fiction movie. So very much appreciate your work as well. So thanks for having us.

Robert Niichel

Yeah, thank you for coming on today.

Sacha Heppell

And how can potential partners contact you?

Artur Adib, PhD

Yeah, absolutely. So the best way is to just email me directly for now at: Artur@biocogniv.com

Sacha Heppell

Okay, awesome. Thanks so much for spending the time with us.

We look forward to staying up to date on your progress. I appreciate your time.

Learn more about Biocogniv: 

https://www.biocogniv.com/

 

PodcastCoverForFeature

WHO WOULD HAVE THOUGHT

Listen and Subscribe to the Podcast

The perspectives of digital health thought leaders today and an inquiry into digital technologies of tomorrow.

 

Listen Now