What is NVIDIA actually doing in healthcare? (David Niewolny)

[00:00:00] Dear listeners, welcome to Faces of Digital Health, a podcast about digital health and how healthcare systems around the world adopt technology with me, Tjasa Zajc.

[00:00:14] NVIDIA is best known for GPUs, but its healthcare strategy is a full stack. Compute to train models, compute to run them as the point of care, and simulation to generate the data that medical robotics lacks. In today's episode, you will hear from David Niewolny, Senior Director and Global Head of Business Development for Healthcare and Medical at NVIDIA,

[00:00:42] who explained how that stack that I just described underpins autonomous x-rays and ultrasound, the work that NVIDIA has been doing with GE Healthcare, the impact NVIDIA wants to have in surgical robotic simulation, and more. David argues that open models are a regulatory necessity in healthcare, that software-defined medical devices will shrink innovation cycles from years to months,

[00:01:10] and that the path to autonomy in surgery will follow the one taken by autonomous vehicles one level at a time. We also discussed whether AI efficiency turns into clinical burnout, whether synthetic data reflects real patient populations, and what could still stall adoption. Enjoy the show, and if you haven't yet, check out our newsletter, which you can find at fodh.substack.com.

[00:01:39] That's fodh.substack.com. And if you will enjoy the show, make sure to leave a rating or a review wherever you listen to your shows. It only takes a second, either in iTunes if you listen to the show on an iPhone, or if you listen to the show in Spotify. There's a very easy way inside the episode to make sure that you rate this episode. Any feedback is welcome.

[00:02:07] And with that, let's dive in today's episode. David, hi, and thank you so much for joining me on Faces of Digital Health to talk a little bit about what NVIDIA is doing in healthcare,

[00:02:37] and how do you see healthcare AI and just healthcare in general moving forward with all the technology that's now become an everyday part of healthcare delivery. You are the Senior Director and Global Head of Business Development for Healthcare and Medical, and I'm happy to have you here. I'm really excited to be here and love all the great work you're doing to really evangelize all these digital health applications.

[00:03:03] And I think as you clearly articulated, I think accelerated computing and NVIDIA are becoming a foundational piece to a lot of those components. So excited to share a little bit more today. Yeah, absolutely. So let's dig straight into it. As you said yourself, NVIDIA is known to the general public by its GPU capabilities and market-leading position in this space.

[00:03:26] And at the same time, the company has been present in healthcare for several years with different approaches, different solutions related to scientific software, model development, agent simulation. Robotics is becoming more and more interesting. So for starters, can you give us an overview of everything that you are doing?

[00:03:50] And can you describe a few examples of how these products work and what kind of companies or healthcare providers use them? So we have a little bit of a broader overview. Yeah, I'd be happy to kind of open with some context. I think as I mentioned, you look at AI and robotics, and I think it's pervasive across a number of different areas of healthcare. An area that I'm specifically focused on, healthcare and medtech, I think our work actually expands even beyond that.

[00:04:18] You're seeing a lot of work even in the area of life sciences and digital biology. So if you look at the way NVIDIA is kind of breaking up the market, we do see applications of AI and robotics in healthcare delivery. We see it in medtech organizations, and then we see that in a lot of the life sciences organizations anchored in some of the large pharma applications.

[00:04:41] And if you look at it across the board, NVIDIA's entire goal is to accelerate the adoption and eventual deployment of these AI-enabled robotics applications for the betterment of healthcare. And really that's focusing around operations. It's focused around clinical care. It's focused around improving both the patient and clinician experience.

[00:05:06] And we're really doing that by providing all of the foundational tools that our ecosystem of developers can begin developing upon. And as I mentioned, that's largely digital health organizations, medtech organizations, an entire ISV ecosystem that is building the foundational platforms that digital biology is being built upon today. And we do that by providing specific tools and frameworks.

[00:05:36] And you can see from the healthcare and medtech perspective, we provided a full stack computing architecture or training, largely anchored in a lot of our work with open models and open model development. We provided that foundational stack in terms of edge deployment because a lot of this AI and robotics needs to be deployed at the point of care. We've done that with a full stack.

[00:06:28] Which is really a full stack. Which is really a complete set of agent tools necessary for the development of next generation pharmaceuticals and different molecular compounds. So we're really trying to create these full stack computing architectures for the developers to begin building and then eventually deploying a lot of these new AI and robotics applications. And I think a couple examples that I can share specifically on the medtech side.

[00:06:58] Last year, we put together a very large collaboration with GE Healthcare to really drive forward a huge efficiency change in X-ray and ultrasound applications. And a lot of that is focused around how do we make those procedures more accessible to more people. And in close collaboration with GE Healthcare, we identified this path towards autonomy.

[00:07:22] How do we begin making these X-ray and ultrasound, which are the most widely used imaging applications, much easier and accessible to more people. Frankly, by scaling the individuals that are providing that level of care. One example, the GE Healthcare moving towards autonomous X-ray and ultrasound.

[00:07:41] We've worked closely with Johnson & Johnson MedTech in terms of helping them use simulation to augment their data sets to begin increasing the deployment on their Monarch robotic platform. Specifically for urology, this entire use case is driven by creating models that can now put us down a path towards making those procedures more autonomous.

[00:08:10] And then if you look at just a couple other examples we've used, we've collaborated with Open Evidence on adopting our Nemotron platform to help build agents. We worked with SORD Health to also adopt an Nemotron platform to create AI-powered sessions that they can actually begin working with patients even outside of the clinical atmosphere.

[00:08:33] So we're really an enabling technology that helps all of these companies accelerate their AI and robotics development. If we unpack some of the things that you mentioned, what do you refer to and how do you see the importance of open source and open models? What does that mean in your particular case? I think you've probably seen it across the board. NVIDIA has been a huge proponent of open source and open models.

[00:09:03] And I think that's especially important when you start talking about healthcare and medtech applications. You look at healthcare and medtech, I think first and foremost, highly regulated industries. Everything needs to be explainable. And if you look at a lot of the great work that the Frontier Labs are doing, a lot of the models that they're developing are generally proprietary. They don't necessarily completely provide all of the data sets that they were trained off of.

[00:09:32] They definitely don't provide the open weights and weights of those models. Where the open source community, specifically for an application like healthcare is so unbelievably important. Specifically from a regulatory perspective, being able to provide exactly the information to be able to explain how these models were created. I think a secondary piece to the benefits of open models is healthcare data is specifically unique.

[00:10:02] We see these open models as being something that can be used and fine-tuned for specific populations. So it really gives the entire ecosystem the ability to start with the foundation and then create differentiated products based on these open models that are unique to the specific challenge that they are overall developing. So at the end of the day, I think open models, as we are seeing, I think are going to continue to become more and more pervasive across a number of industries.

[00:10:31] But you can also see the importance of an open model ecosystem when it comes to something that is highly regulated like the healthcare industry. Can you go into the autonomous radiology and imaging that you mentioned before? How does that look like in practice? Is it just enabling radiologists to do more in an hour?

[00:10:54] Or are we actually talking about self-driving systems that do not need a human present to do the imaging? I think first and foremost, I'm going to start with both, but all anchored in being that healthcare is probably one of the best personal industries that any of us can interact with. And from that aspect, we see AI and robotics as something that is always going to augment a clinician.

[00:11:22] It's going to make them far more efficient. Essentially, the word I like to use is giving them superpowers that they hadn't had before. They can now see more patients, which improves access to care. They can actually begin doing their work much more efficiently. So we do not really see this moving towards a path where we're going towards full autonomy. It's augmented autonomy. Really, you're always going to have a clinician in the loop in these cases.

[00:11:52] But now breaking it up into the two applications that you mentioned, and this is why I said both. If you look at the GE example, in that particular case, we're looking at a ultrasound technician, which, again, there's not enough of them to see the patient population that needs ultrasound. How can we allow that ultrasound technician to perform more ultrasounds much faster?

[00:12:19] And with that, that's where this path towards autonomy comes into play. You might have one ultrasound tech now actually setting up but working on four rooms at once versus just a single room. How do you actually begin taking things like X-ray, which is the most widely used imaging diagnosis platform, and scaling that to places where they might not actually have an X-ray tech?

[00:12:48] And this really comes to leveraging the full stack of NVIDIA technology. You can almost think of an AI digital agent that's actually helping direct check-in. You can see that agent actually helping position the patient and talking them through where they need to be positioned. You can see that X-ray now operating essentially as a robot and positioning itself based on the patient position to get the best possible image.

[00:13:17] You can now also see agents running in the background that are analyzing that image, making sure that we have the right image to be able to send to the radiologist. And now if you take that one level back further, you now have a number of different radiology software companies that are doing analysis of that image and providing reports on that radiologist around that image in the form of a radiology report.

[00:13:45] That essentially doesn't make a specific diagnosis, but actually makes suggestions to the radiologist in terms of what they should be looking for, making sure that they didn't miss anything. We have a number of different companies that are focused specifically in that area with the work we've done with YAC being one of those. So there's a number of different areas that we're working in this space. And when we talk about autonomy, a lot of this is, thank God, semi-autonomy.

[00:14:15] We're not going to a path where we're removing humans. We're just augmenting them with intelligence that they didn't have before. There's definitely a huge need in healthcare to somehow increase efficiency because based on the workforce shortages, it's impossible to just create the workforce that would take care of the rising needs that we see.

[00:14:34] But I do wonder when we talk about the augmentation of work and higher efficiency and clinicians being able to do more, how far do you think that can go? And what I mean by that is that one of the discussions that it's starting to open up in the conferences or discussions on AI is this cognitive overload that people are experiencing.

[00:15:00] The fairy tales of doctors having more time for patients are over. No, we now know that they just need to do more and see more people and just make more decisions. And for humans, it's exhausting. Even if you, it's exhausting to be in the loop to a certain degree.

[00:15:18] How do we mitigate the burden that could potentially lead to a new wave of burnout of clinicians being primed to use AI and all the tools that support their work? I think hopefully it's coming through as part of some of this discussion. That is specifically the challenge that we're looking to solve with a lot of these AI tools.

[00:15:46] It's removing that cognitive burden and in many cases, removing a lot of the busy work that a lot of these clinicians are doing. If you look at some of the examples in terms of clinical documentation, I think it's a publicly available story that we've been working very closely with Abridge. You look specifically at clinical documentation. Most of these healthcare providers are doing that at the end of the day.

[00:16:14] They're seeing patient after patient and then having to review their notes and then make a clinical document of what was actually the outcome? What did they talk about during that session? And I think clinical documentation is one of those clearest examples of making healthcare more efficient. You're now actually breaking down each of those visits at the time of the visit, pulling out the most important information. But as I mentioned, this isn't fully autonomous.

[00:16:40] This isn't actually going to an electronic medical record or going back to the patient or making clinical diagnosis. It's making recommendations that all need to go back and be reviewed by that specific clinician. And you look at the amount of time, effort, energy that it's saving that clinician from having to do one at the end of the day, which theoretically should reduce their cognitive load and burnout.

[00:17:07] And at the same time, doing a much better job at documenting at the time of care, as opposed to doing that potentially eight or 12 hours after each of those visits. So if you look at a lot of the foundational technologies that we're building, it's really designed to work with our partner ecosystem to develop solutions that are going to ease that cognitive burden on the clinicians and eventually begin moving towards better patient outcomes.

[00:17:35] Among all the solutions and providers that you are supporting or are using your technology, do you have a favorite, the favorite application, something that you're most excited about? I don't know. I think it's really interesting. NVIDIA, if you look at the way we work across the foundational ecosystem, it's very hard to choose favorites. There's a lot of really great applications. There's a lot of really great companies.

[00:18:01] And we are truly collaborating with as many of those as possible who are looking to move as fast as we possibly can. It might actually be a good time just to explain what a good NVIDIA partner looks like. And really what we're looking for is what is a customer that has essentially an exceptionally hard problem

[00:18:25] that we see as clinically and from healthcare as a whole has the opportunity to make a huge impact. And that's really what brings the two companies together. And then foundationally, we end up working, looking at that problem and seeing how can NVIDIA work closely with that particular partner to accelerate that development, to make that vision that we both have a reality much faster.

[00:18:53] I mean, both essentially need to have unique capabilities, largely NVIDIA's capabilities on the technical side. And our partners, med tech companies, hospital providers, biotech companies are really bringing that clinical and biological expertise. When you bring that together, it really helps accelerate a lot of these products to market much, much faster. You have the leader in their clinical domain.

[00:19:18] You have the leader in the technical domain collaborating to make something that's really going to foundationally change the way healthcare is delivered. And now if you start looking at some of those great applications, I already highlighted the GE application, which is really exciting. I think the area that we see the most promise when you start looking five and 10 years out is this whole idea of robotics and augmenting a workforce with robotics.

[00:19:46] I think one of the stats that we mentioned recently is just, you look today, I think 20% of all surgical operations are being done by surgical robots. And that is moving very fast, increasing at a very fast rate. And the ability to scale that is largely based on the small amount of robotic surgeons that are available. So how do you actually democratize that technology?

[00:20:16] You make it much easier for those surgical robots to operate. And this is really a path that we're taking down and we've taken a lot of learnings from other industries. One being autonomous vehicles, which NVIDIA is a leader in developing and deploying these autonomous vehicle solutions through our partner ecosystem.

[00:20:37] So at the start of this conversation, I think I talked a little bit about the whole idea of a full stack compute for model development, one for model deployment, and then one for creating simulated data sets for us to begin training these models. And we're working directly with a lot of models. And we're working directly with a lot of motors.

[00:21:03] path for these surgical robots. And we see it moving in the exact same way we saw autonomous vehicles. It's not, you go from a driving a car to a full self-driving car. What you end up seeing is you take steps. There's different levels of autonomy. In the automotive world, we moved into lane change warnings, automatic braking. We're looking at doing that same sort of contextual

[00:21:30] awareness and action, specifically in the areas of robotic surgery. And that's really going to make robotic surgery that much more safe. But it's also, I think, more important to get to make it much more accessible. And I think you're going to see in these surgical robotics applications, which have a lot of data that show essentially they're safer, patients recover faster, much less invasive on surgical procedures. And I think that's one area that we're going to see grow. And then expanding on

[00:22:00] that robotics component. If you also look, there is obviously a huge staffing shortage in hospitals and healthcare organizations. There's a lot of work that's being done in a hospital that is not necessarily patient-facing. And there's a huge opportunity for using robots to help automate a lot of these non-patient-facing type roles. So think of more of the operations behind the scenes of getting rooms ready,

[00:22:29] cleaning rooms and others. We're working with a lot of the leaders specifically in those spaces to begin educating them on how they began deploying robotics to provide a much more efficient healthcare experience. What do you see has changed in robotics specifically, say, in the last 10-15 years? I'm just thinking from the perspective that robots that robots are used in different ways in healthcare. For example, they can also be

[00:22:58] used in pharmacies. But if you just take the global perspective, they're not necessarily widely implemented because they cost a lot. So how does the cost of robots and maybe even these surgical robots, how does that impact? How much can the technology scale to have an impact? No, I think it's a great call-out. And I think in the past, you're correct. A lot of these robotic

[00:23:27] solutions have been exceptionally expensive, which makes them not as widely adopted. What you have really seen in the last, frankly, two and three years is a huge influx of robotics development. And with that, that has largely been driven by a lot of the capabilities and tools that NVIDIA has built,

[00:23:51] as I mentioned about those specific industry-specific stacks on training, simulation, and deployment, that's now making it much more accessible for a partner ecosystem to begin building these solutions, which now creates a much more competitive market. Which in much more competitive market, competition ends up driving, I think, one better products, but also creates a much more cost-competitive

[00:24:17] atmosphere. And I think that's what we're seeing generally across the board. So one, it's much easier to build these products. So the cost to build them is going down. Because it's easier to build them, you're seeing more players in the market. When you have more players in the market, you're seeing competition, which is actually helping drive the cost down, which is now allowing all of these healthcare providers to begin looking at this from a true ROI perspective. And I think that's

[00:24:45] one of the, I don't want to say biggest changes, I think deploying these solutions has always been around, can you show the return on investment? And I think we're now getting to a point where that return on investment is becoming clearer and clearer every day. So in addition to providing better patient care, better patient outcomes, you're also now hitting price points that are making these, the return on the investment from an operations perspective, much more clear.

[00:25:13] I guess that's a welcome news, especially since when it comes to AI and ROI, it's a tough question. Many hospital leaders struggle with, I don't know what's your insight into the market in that sense. I think, I don't want to say it's necessarily a change. I just think the whole idea of return on investment is becoming much more important, specifically in terms of selling the value that

[00:25:39] our partners are bringing to market. I think there's definitely a patient care, and there'll all be a patient care component to all of these applications. But I think as you clearly articulated, if you improve patient care, at what cost can these healthcare organizations absorb? And I think what you're seeing is now a lot of that narrative is changing in terms of we're improving patient care,

[00:26:05] but in addition to that, from an operations perspective, we're actually able to allow you to see X number more patients or lower your operating costs by X percent. That's becoming much more of a value conversation that I frankly don't think was highlighted as much when you look five and 10 years ago. Yeah, maybe, I guess maybe there wasn't that much data available yet to be able to make those claims.

[00:26:30] But that's just a guess. When you mentioned simulation and synthetic data, can you talk a little bit about how do you create those datasets? And what I'm aiming to understand is, with all the variations in guidelines and care delivery, how can these solutions or this simulated data be applied to a particular hospital that has a particular population that they take care of?

[00:27:01] Yeah, I can definitely go into that. I think the first and foremost is defining why we see simulation and simulated data as being so important. I think first and foremost, you look at what the Frontier Labs have done in terms of large language models and their development, and frankly, the sheer amount of data that they have.

[00:27:21] I think the number that we've seen quoted is it takes almost 2 billion hours of data to begin actually creating a good large language model. You start moving into the world of just robotics, and you just don't have that single amount of data of the world. I think in general purpose robotics, they say there's 10,000 hours of data.

[00:27:45] And then if you break that down into surgical robotics or some of these healthcare-specific robotic examples, the amount of data is frankly very minimal, specifically when you get into corner cases and specific use cases that you need to train these models. And that was really what drove us to begin creating these simulation platforms. And from an NVIDIA perspective, we've created a world foundation model called Cosmos.

[00:28:11] And then what our healthcare business unit does is then essentially fine-tune that. And we've created a product called Cosmos H, which is part of our Isaac for Healthcare simulation platform that is very specific to certain use cases. First variation of Cosmos H is specifically trained on surgical video as we move down this path towards autonomy and surgical robots.

[00:28:36] And the way we actually capture that data and begin creating this simulated data is by creating our own open source. One of the initiatives that we have internally at NVIDIA in alignment with the discussion we had earlier on open models, open datasets, we introduced an initiative called OpenAge, which now allows all the leading healthcare organizations, leading med tech companies to

[00:29:02] contribute some of their data to create what is now the world's largest dataset for surgical robotics. We then take that dataset and begin training it to create a world foundation model and a specific simulation platform to create data that we can then augment to close that data gap that I articulated

[00:29:27] between what the large language model Frontier Labs are creating and what we have available for things like surgical robotics. So essentially, we assemble large datasets from our standpoint, we call that OpenAge. And then we take that, train it, create simulation platforms that we use to augment to data, and then we use our own

[00:29:51] training infrastructure to then build these open models that the entire community can then build and develop on. And that's what we see is accelerating this robotic development. And we're starting to expand that OpenAge initiative past surgical data into ambient data for the OR, ambient data for the hospital, and other different use cases where we see simulation and simulated data being exceptionally important.

[00:30:18] Access to data is in general a really tough part of the development to tackle if you're a healthcare technology company. I know you're based in the US, but given that NVIDIA is working globally, how does that access to data

[00:30:40] transfer to what you can do or can't do? Can you also create simulated data for other markets? How sure can you be that simulated data reflects and is a comparable digital twin to the real world? I think there's multiple steps to this. I think the challenge you articulated is exactly why we're trying to create these open models.

[00:31:06] We see each organization has a small amount of data, data that's good enough to fine tune a model, but not large enough to build them all. So if you act and go through that overall kind of process that we put together, where we gather inputs from the entire industry to build the largest data set we can,

[00:31:32] use that to train the simulation model, which then provides us access to a large amount of data that we can then begin training all of these large open models based on. So essentially what we're working to do with open models is solve that data gap challenge. We realize each organization has enough data to fine tune, but not to train the initial foundation model.

[00:32:02] And that's really what we're looking to build, which accelerates the entire industry. And that open H platform that we put together is really what our goal is to create the world's largest data set in each of these specific domains to give us the tools and data we need to create the best possible simulated data. And to answer your kind of final question on that was how can you actually guarantee that simulated data is good enough to begin doing that training?

[00:32:30] And frankly, that's the output of the model. If you actually look at the output of the model, you need to do verification and validation on that specific model. And that is really where you start seeing is the simulated data good enough to begin training these models and the evidence that we have so far both internally and with our partners leveraging this overall process for generating simulated data, fine tuning models, and then creating specific robotic policies or AI applications.

[00:33:00] We're seeing some very good results in terms of correlation. How do you maintain quality in this whole picture? I'm just thinking from the general challenges that we have with AI, such as data drift, such as AI model drift. Now, like there's the reality, there's the simulated data, and then you're also doing training and practice on a simulated data. So it's like on the third layer, you're trying to get to something that's similar to reality.

[00:33:29] So I don't know how do you see that balance between thinking that your data is good and being sure that the data and the simulations are what is also happening in real world? I think a lot of it just comes from very similar to what the medical device and medtech industry has had for a number of years. There's specific requirements in terms of verification and validation.

[00:33:56] And if you think of AI applications and AI models, these are no different than medical devices in many cases. There needs to be and is being built a governance model within these regulatory agencies at these medtech organizations to ensure that they have checks for things like model drift over time. And that becomes especially important when you get into areas of reinforced learning.

[00:34:21] You want all of these models to continue taking in new data and essentially retraining itself to make sure that it is operating. But with that, there is a level of governance that is being established within each of these companies building and deploying these AI and robotics applications to ensure that they are actively managing and monitoring things like model drift and the overall execution and accuracy of the AI tools that they have.

[00:34:51] Yeah, governance is a popular word in the healthcare data space and especially with AI. I do wonder how do you apply what does good governance look like when it comes to multi-specialty agent orchestration? So how do you look at best practices in the governance of AI agents in healthcare?

[00:35:17] I'd say you look at governance as a whole and I think we kind of look at it in different layers. It starts with bounded tools. So making sure that all of the tools which NVIDIA provides are being essentially developed in a way that makes sure they're safe, secure, and validated. Next piece is ensuring you're actually controlling data access.

[00:35:43] And then the last piece, as I mentioned, you always need to have in all these cases a clear human decision point. A lot of these AI applications are going to have a clear human decision point in there. All of this is just designed, as I articulated earlier, to give these clinicians essentially superpowers to begin doing more as opposed to removing any of these decisions directly from the clinicians.

[00:36:11] Are you worried in any way about the potential of having superintelligence in healthcare? With the discussions the last week, it would be silly of me not to ask because in healthcare we're so focused on the control, on human in the loop. And here we are discussing how the world is going to end because somebody is going to take over and it's not a human. I think it's an interesting discussion.

[00:36:37] I would say if you look at healthcare as a whole, it has moved slowly and it has moved slowly for a reason. And I think a lot of that is we have a lot of the regulatory agencies, the controls in place to not allow technology to be deployed without the right amount of rigor, verification, and validation.

[00:37:06] I think what NVIDIA has provided is a whole set of tools that allow all of these companies to move faster in terms of their development. That the next piece of this will be how do we verify and validate that these applications, as I mentioned earlier, are showing that return on investment and are actually making a positive impact to patient care.

[00:37:30] That I'm frankly of other areas and other industries less worried about the idea of superintelligence moving too fast in healthcare.

[00:37:39] I think healthcare is going to do a great job in terms of regulating itself, in terms of using these AI models and robotic applications where they make sense, but where they are justifiably used, where they can truly have proof points of improving patient care, improving access to care, as opposed to nonchalantly deploying these models without the appropriate regulatory verification, validation, and governance.

[00:38:07] Do you notice a lot of disconnect between what is technically possible or what are the ideas of where AI is going and what's happening on the ground? So where do expectations still outpace the reality at the moment? I'd say in many cases, the expectations really just end up being around a timeline perspective.

[00:38:33] I'd say for the first time in my career, working in technology starting back in the early 2000s, where it's almost like this vision of collecting data from medical devices to begin driving intelligence that can impact clinical care. We now have all of the tools that are essentially available to help companies begin building these solutions.

[00:39:01] And if you can say the technology is now there, it's really going to be about the execution of organizations that are forward-looking, taking this technology and building it, because now they have the ability to build faster than they ever have before. Medtech organizations in the past have generally built everything from the ground up. They're now much more open to partnerships in terms of helping NVIDIA do some of the undifferentiated heavy lifting and then executing on the clinical components to that.

[00:39:30] But I do think the biggest piece to this now and the biggest challenge is going to be the overall change management within healthcare industry as a whole. How willing are the hospitals to begin, one, piloting this and then testing it, showing the return on investment, and then deploying it widely at scale?

[00:39:51] There is not only a technology component, but there's also a change management component that I think will take some time to work through. But I think you're going to see leading organizations, both on the medtech side as well as the healthcare provider side, adopting these technologies, beginning to show pilots, beginning to show success, which I made the link earlier, is not dissimilar from what we've seen in a number of other industries.

[00:40:17] And the one we follow most closely is, because it is regulated, is the autonomous vehicle industry and how that kind of moves from both the technology perspective, but also a change management. You can see we're all getting much, much more comfortable with the idea of autonomous vehicles on the road. We see that same sort of path happening within a lot of these healthcare AI and robotics allocations.

[00:40:42] But to your point and to the point of change management, people are always resistant to change, especially when it comes to workflow. So humans will make sure that the speed is not too fast. Correct. You mentioned how different it is for medtech companies to develop their own products.

[00:41:07] How does the speed of what's possible because of the computation power and the development AI, how does that go together with the challenge of hardware development in healthcare and all the processes that go under that when you're trying to make sure that your device is certified? So it's one thing to do an update on a software if you just provide the software.

[00:41:32] But if you provide hardware and on top of that, a software inside the hardware, like the timelines of how fast you can go to market, how fast you can change something are much different. So given that you work with medtech companies, what are you observing in that sense? I think one of the most exciting areas of development is this idea of software-defined medical devices.

[00:41:56] I think just as you mentioned, the speed of building a hardware product and then building software on top of it, taking it through a regulatory approval has been one of the things that has really gated a lot of innovation in healthcare and medtech development. That process, as you're well aware, is exceptionally long, anywhere from three, sometimes upwards of seven years.

[00:42:24] So you have this long timeline of development. I think what we're starting to see a lot of these medtech companies now move and support is the software-defined methodology, where, yes, you might spend a good amount of time in terms of reinventing and redeveloping a hardware application.

[00:42:44] But the idea of actually building and deploying software on top of that, that creates new capabilities on that same piece of hardware that still will have potentially a 5-, 10-, 15-year lifetime. It's basically taking the hardware component and treating it as a sensor. And now all of the intelligence on top of that is being deployed via software.

[00:43:09] And we know that the cycle and speed of software and software updates can happen much faster. So no longer is this a 3- to 5-year development. This is now more of a 6- to 12-month development. And I think we've even seen a lot of the regulatory agencies begin supporting this.

[00:43:29] And I think one of the clearest views of that is the PC that the FDA has, the pre-change procedure, pre-change control procedure plan, where you now specifically design for these AI applications. You can essentially document your process and procedure for making these regular software updates, which does include how are you going to verify and validate this software prior to then deploying it onto these hardware devices.

[00:43:58] It's that, as I mentioned, they will continue to live for a long period of time in each of their application areas, but they're going to continue getting smarter via the software that's deployed on top of them. And that layer of intelligence should become increasing because that's really happening in a 6- to 12-month increment versus currently these med device refresh cycles are more in the areas of 5- to 10 years.

[00:44:23] It sounds like a much-needed change in the way that we regulate anything that goes on the healthcare market to make sure that it doesn't stall too much because of regulation. So while we're touching the regulation in the last few years, what are your observations in terms of how is regulation adapting well, where it should adapt even more?

[00:44:49] We always have this challenge of regulators not being the tech-heavy individuals that are in the industry, that are on the cusp of development and know what are all the latest technologies that are out there. And yet, there you are, somebody stepping behind the development, trying to figure out how to regulate it. It's an impossible task. I guess I don't want to build it up to be an impossible task. It is a challenge.

[00:45:17] Anytime you're doing something new, there's education that needs to happen. And I think the great news is we're multiple years into this overall education process, specifically in that area of software-defined devices. And you're seeing more and more regulatory actions in terms of different regulatory processes and procedures that are allowing these organizations to begin the education process of the regulatory agencies.

[00:45:46] And I know in many cases, these regulatory agencies are built out as someone that's slowing innovation. But I'd say we have even seen a shift with the regulatory agencies in terms of this is coming. This has a huge opportunity for improving patient care. The time is now to begin improving efficiencies and improving access to care that a lot of our MedTech partners are working very closely with the regulatory agencies in terms of that education.

[00:46:13] And if you think about the regulatory agencies, the number one priority and the reason they exist is essentially around keeping us all safe. They don't want this industry to move at a pace that is reckless. And you end up having issues where you have recalls and others. You kind of anchor that with the idea of patient safety is the main concern.

[00:46:34] It's really the burden of proof comes on the MedTech companies of ensuring that what they're providing is safe and effective to be deployed at scale. And what are those things that need to be put in place to do that? And I can tell you that our MedTech organizations are working with all of the global readiness or agencies to ensure that we're going to find that happy medium that still indexes on speed,

[00:47:03] but is safely anchored in the idea of safety and effectiveness of these solutions. Absolutely. Yeah. Patient safety should be at the core, especially when we look at the whole care delivery from the patient perspective. Yeah. We are coming to the end of this discussion. So maybe as a wrap up, you've been in the industry for over 18 years.

[00:47:31] What are some of the things that excite you most? How do you see the best case scenario of where health care could be realistically more than enough challenges to not say that everybody is going to be healthy and we're not going to need health care anymore because everything's going to be automated and efficient. So what's kind of something that is getting you excited to go to work every day? There's so many things that make me excited to wake up and go to work every day,

[00:48:01] but you highlighted a couple of things earlier and I'll hit on these points again. But throughout the course of my career, it really started with this entire vision of the internet exploded in the early 2000s. And it was, what if you could actually begin gathering data from all of these medical devices and what are the insights you could begin deriving from that?

[00:48:23] It's frankly amazing that it's taken us 20 years to get here, but we are now at a point where from a technology perspective, we have access to the data, we have interoperability tools that give us the ability to non-normalize that data. We have the accelerating compute infrastructure to begin creating these AI models that were not possible even three years ago.

[00:48:48] And then you have companies like NVIDIA that have the specific focus on accelerating AI and robotics applications for a specific domain of health care. That I am confident that this is the first time in history that from a technology perspective, we have all of the pieces and the right players focus on the right things. So the tooling is there, the technology is there.

[00:49:16] The next piece is, do you have the appetite of the MedTech organizations and health care as a whole to begin adopting these technologies? And it's unfortunate that in many places around the world, health care is in the state that it's in. We have the staffing shortages. We have increasing costs that now you have.

[00:49:38] The time is right to begin leveraging all of these AI automation and robotics applications to improve health care. So you have timing, you have the technology that for me, this is really going to be the golden age of MedTech and health care technology development, because you now have the entire ecosystem working together to solve what is truly a global problem with health care development and deployment. So I'm excited. So I'm excited.

[00:50:06] One, you have the technology, you have the desire, and people are actually moving and activating to solve what is arguably probably the biggest opportunity for AI and robotics as I see it. I will ask you one more thing. What do you think could be the biggest challenge for this opportunity to be seized? And the reason I'm asking that is because you basically said that all the pieces have fallen into place.

[00:50:36] But one of the main discussions that we're now seeing in Europe when it comes to digital transformation is that COVID is far behind us. And the budgets, because especially public budgets are moving from health care to defense. So it's much more difficult. It's getting more difficult to just keep investing in all the technologies. So I don't know if you have an opinion on that. And what are potentially other challenges that you see that might impact the speed of progress?

[00:51:05] I'm not going to go back and dwell on the piece of change management. I think change management is one component that I think both NVIDIA as well as the entire MedTech ecosystem is going to continue educating clinicians, healthcare providers, essentially the entire industry on this is safe, effective, and has some significant opportunities to improve clinical efficiency.

[00:51:30] The second piece really ends up coming around that return on investment. And I think that is probably the most important discussion. It's yes, we have the technology. Yes, the time is now. But if it's not actually improving the efficiency of a healthcare organization, yes, that could be a sticking point.

[00:51:53] But I think every one of these companies sees that and their business relies on creating solutions that achieve that return on investment. And that's no different than any other industry out there. Nobody adopts hardware, software, any sort of technology unless it's actually improving their business. And in this particular case, you look at some of the business metrics and it's patient outcomes. It's improving access to care.

[00:52:22] How can you treat more patients? And then obviously the bottom line of improving the hospital operations. So I think those are things that the entire industry is going to continue driving towards. And it's really the burden of that becomes on the medtech organizations to begin creating these businesses that show that clear ROI to the entire industry.

[00:53:09] Stay tuned.