Latest / StartUp Health NOW Podcast / GE Healthcare’s Karley Yoder Builds Bridges With AI

Transcript

GE Healthcare’s Karley Yoder Builds Bridges With AI · StartUp Health NOW Podcast

• 0:00
  [on-hold music] What my personal, um, mission is in healthcare is to always be at the forefront of the technology that, um, is most powerful in the day, but then to apply it back to solving healthcare problems that matter.
• 0:23
  [upbeat music] Welcome to Startup Health Now, the podcast where we celebrate the innovative entrepreneurs and investors reimagining health. I'm Logan Plaster, editor-in-chief at Startup Health.
• 0:40
  Carly Yoder, my guest on this episode, has had an impressive career in health innovation. Her career took her from Apple to Doctor on Demand, and now she's at GE Healthcare as VP and GM of Artificial Intelligence.
• 0:53
  There at GE, Carly's work touches nearly every department as the company looks to integrate smarter AI into all that it does. But she's not just strong on the tech.
• 1:03
  She's also made a name for herself in understanding the human side of using AI to improve health. What are the ethical implications? What are the limitations?
• 1:12
  In our interview, we'll get into some of those interesting issues, plus we'll talk about how Carly assesses opportunities for partnership in the startup ecosystem.
• 1:21
  This episode was taken from a recent fireside chat in front of a live audience of founders from the Startup Health portfolio, so you'll get to hear some of their questions as well.
• 1:30
  As always, if you enjoy this podcast, make sure you follow us on your favorite podcast player so you don't miss an episode.
• 1:36
  Also, if you're an accredited investor and wanna learn about our Health Moonshot Rolling Fund, which we recently launched on the AngelList platform, go to healthmoonshots.com.
• 1:46
  Now, onto our conversation with Carly Yoder of GE Healthcare. [upbeat music] Uh, I'm excited about this conversation, um, Carly.
• 2:00
  You said recently in an interview that AI isn't about the technology, it has to be about outcomes.
• 2:07
  Um, this got me thinking, in a way, this is another way to get at our Health Moonshot vision, this idea that health tech can't just be about the tech, about the p- the products and the platforms, it's gotta be about those, um, those moonshots, about making health better for everyone.
• 2:22
  So, uh, I wondered if you could s- talk to us about, you know, what was it about your unique journey in health innovation that got you to that perspective?
• 2:32
  Um, and maybe you could elaborate kind of what you meant by that, uh, that concept. Sure, Logan.
• 2:39
  So it might seem like a strange thing for an AI leader to say, but I actually think if I do my job right, it won't exist five years from now, um, because we, we don't have, uh, GMs of web enablement, as an example.
• 2:52
  So y- [laughs] you asked two questions, Logan. You asked what got me to this place, um, and then can I elaborate a little more on what I meant by the statement. So, so let me, let me address those in order.
• 3:02
  So, um, what got me to this place? Look, my, my entire career and, uh, passion in healthcare is around being an effective bridger. Um, and so what I mean by that is my background's biomedical engineering.
• 3:17
  My entire career has been kind of in this product management, uh, business translation space.
• 3:24
  And as I look back at my journey, um, when I think about early in my career when I worked at, uh, Duke on, you know, clinical trial translation, at Stanford on MRI development, um, what made me good in those jobs was not that I was the smartest in the technology.
• 3:41
  In fact, I was learning the math when, uh, I was trying to help with the physics around MR, right?
• 3:46
  But what made me effective was the [laughs] ability to understand enough to translate to what it meant to a science community or to a c- a customer base that wanted to grapple with that. Right?
• 3:58
  As I think about the work I did with health insurance exchanges.
• 4:01
  So early in my career, um, I had the opportunity to work across, uh, seven or eight state governments up and down the East Coast as they tried [laughs] to respond to this small thing called the Affordable Care Act, and what they needed to do to transition and set up state-level health insurance exchanges.
• 4:19
  Um, this was a bridger role that required the intersection of technology, policy, business, and, and let me tell you, becoming a Medicaid director, um, doesn't happen because you're brilliant at running system integrator programs.
• 4:33
  And so I had the opportunity to work at a company that did that type of work.
• 4:38
  Um, and the skill set that was needed, more than a deep, deep, deep expertise in any one function, was the ability to be deep enough to connect groups that didn't speak fluently to each other.
• 4:49
  Um, at Doctor on Demand, similar. I, I actually joined-- I worked there very early in their telemedicine journey. This is, this is a company probably many of you fam- are familiar with at this space.
• 5:00
  Um, but, but way before telemedicine, um, [tsks] got the acceleration we've seen recently, helping them think through, how do we shift from a solely B2B market strat- or B2C into B2B, right?
• 5:13
  The, the, the strategy when I joined was, was very much, let's get on a daytime television show. Let's try to get that one person to call in, as opposed to thinking about, how do we go to employers?
• 5:21
  How do we go to payers? How do we go to self-insured businesses, et cetera? Um, similar at Apple, thinking about how do you position products in a broader ecosystem. And so I, I think for me, um,
• 5:35
  what my personal, um, mission is in healthcare is to always be at the forefront of the technology that, um, is most powerful in the day, but then to apply it back to solving healthcare problems that matter.
• 5:52
  Um, and it's never going to be about the technology itself, and if you, if you stay stagnant on one technology, you'll fall behind.
• 6:01
  Um, and so, you know, right now, in the last five years, the hypeAround AI, um, has been tremendous. We'll have a chance [laughs] to kind of unpack some of that over the course of this hour.
• 6:13
  Would love questions from you guys on that front.
• 6:15
  But when I think about my role at GE, my job is to harness this AI, not for the sake of the cool technology or the marketing slogan, but to harness it so it drives impacts that matters to our customers and their patients around the globe.
• 6:32
  Better efficiency, better quality, better outcomes.
• 6:35
  At the end of the day, that's what customers will buy, and those are the type of transformations that move healthcare forward, not in, in small ways, but in, in stepwise changes.
• 6:47
  Um, and so AI is a tool, albeit a pretty powerful one, but at the end of the day, it's one tool in a broad arsenal for, for tackling the problems that, that matter in healthcare.
• 6:59
  So that's really what I meant, uh, w- when I said that is, is look, the sooner we get past how a product was built and the sooner we focus on what impact it has for a provider community or a patient community, um, we're going to start talking apples to apples and, and driving impact faster.
• 7:17
  Well, I love that you said that if you do your job well, uh, your position will be unnecessary five years from now. Uh, that's like a, an interesting sort of health moonshot vision for, you know, obsolescence.
• 7:29
  Like if we do, if we do [laughs] we're doing well, run ourselves out of business, uh, in some cases. But, I-- it's a good segue into understanding your current role. What is it that you do as a- Sure... G- GM of AI,
• 7:42
  and kind of what's your focus right now? Yeah. So a- at the highest of levels, two big pillars of, of what I drive in the organization... Actually, let me first take a little step back.
• 7:53
  For those of you who don't know GE Healthcare that well, it's about a $18 billion healthcare business focused in the med tech space, um, with the big portfolios being about $9 to $10 billion in the, the, the big iron imaging, MR, X-ray, CT.
• 8:09
  About another $6 billion in, in ultrasounds or some of the, the smaller device. About another $1 to $2 billion in more of the software solutions.
• 8:18
  Um, uh, specializing, again, in kind of the radiology or cardiology, oncology spaces that leverage images. Um, so, so big, uh, healthcare company that's been around for a while.
• 8:31
  My job, uh, is very much a horizontal that cuts across all of those businesses, and the, the pillars that I'm accountable for and responsible for driving are, number one, how do we pick the right use cases across each of those PNLs that we-- that best leverage...
• 8:50
  And specifically for us, it's often deep learning under artificial intelligence because of the imaging focus.
• 8:55
  But the organic use cases where if we can build a deep learning model appropriately embedded in the devices, it most, uh, differentiates from competition and provides the outcomes that our customers are looking for.
• 9:08
  An example on this organic pillar is what we've done with the critical care suite in X-ray, where the, uh, you know, the best practices, uh, before AI was if you
• 9:20
  came into the emergency room, um, God forbid you were in a car crash, one of the first things that someone would do was they'd roll a mobile X-ray over you. They'd take a scan of your, of your chest region.
• 9:33
  Uh, that scan would go sit in a queue. Hopefully, a radiologist would get to it within 30 minutes. Sometimes it'd be more like several hours.
• 9:41
  If you had a collapsed lung, that-- the window to intervene and do something about it was solely dependent on how fast a human worked through their inbox or work list.
• 9:51
  What we've done with AI is we've trained a model to spot those critical conditions, embedded that intelligence right in the device, and now those, those, those conditions and categories are prioritizing the work list, so you have critical first, not first in, first out.
• 10:07
  That's one example. We're, we're driving that type of thinking across every single med device that GE Healthcare builds. So that's str- that's pillar number one, kind of this organic strategy.
• 10:18
  The inorganic strategy around partnerships is, uh, is an exciting, exciting space as well, uh, i- in the AI world.
• 10:28
  Um, many of you will know these stats, but there's over $4 billion of investment flowing into healthcare startups, and specifically with medical imaging, which is where I, where I, I spend a lot of time.
• 10:41
  There's over 200 startups with a billion dollars of venture capital funding. Um, now, will all of those startups survive five years from now? No. There, there will be some attrition over time.
• 10:53
  But the, the strategy that I- I'm, I'm tasked to drive is how do we partner extremely well with this burgeoning ecosystem in-- of innovation and help take back to our customers a single point of access for all of these different point solutions, um, and a s- and, and almost an invisible way of orchestrating across many, many different, uh, partners so they get one unified experience.
• 11:19
  So at, at the really broad level, uh, what I'm, uh, accountable in my role is the organic AI strategy. How do we put the right deep learning solutions embedded deeply within our devices and products?
• 11:30
  But then driving our, our Edison, Edison's the name of our AI platform, Edison developer program, to work more and more efficiently with the developer community and provide better and better streamlined consolidation of that efficiency and innovation back to our customers so that they can work with, with one partner and, and not hundreds.Um- Gotcha.
• 11:53
  So Logan, if that, that, maybe that went- Yeah, no, it's great... too high or too deep, but. That's, no, that's great. Uh, high and deep is great.
• 11:59
  Um, I wanna get into more of the strategies around those partnerships and collaborations in a minute, but I wanna stay global, uh, for one more question first.
• 12:08
  You mentioned your horizontal view of the application of AI and deep learning across GE. Uh, and given your previous experience as well, you have this breadth of experience, a sort of envi- enviably global view.
• 12:23
  And so to, just to stay broad for a minute, I want to hear from you just kind of where do you see AI heading right now? Kind of what, what are some of your top-line, uh, themes, uh, that you're seeing?
• 12:35
  You just mentioned imaging and the opportunities within imaging. Um, but could you just kinda give us that horizontal broad view for a minute of the s- of where, of you, where you see AI innovation in the market?
• 12:48
  Yeah, great question. Um, I, I think it's, it's pretty general across most areas of healthcare, but let me highlight a couple.
• 12:55
  So, so certainly within imaging, look, deep, deep learning is tailor-made for pattern recognition based on images. Um, some of the other areas, um, around, around d-
• 13:07
  data, so think about monitoring, patient monitoring or EMR data. The, the standardization of that data has to be phase one before the AI can be applied, right?
• 13:17
  One of the beauties of imaging is there's such consistency in how the data is presued, produced.
• 13:24
  In some other areas of healthcare, there needs to be a focused effort to standardize the data so that you can then apply the AI.
• 13:31
  I think, you know, one of the trends that's been accelerated, um, dramatically by, by, by COVID is virtualization. Um, and I think of this in, in two pieces.
• 13:42
  One, clearly we all know the, the, the data and trends around telemedicine.
• 13:47
  Um, y- you know, I, I think one of the, the most, the recent stats I, I saw was that, um, you know, there was a projection that we'd be at, what, I think 20% of our clinical encounters in 2020 driven by telemedicine.
• 14:03
  I saw that same study projected 70% by 2023, which I, I, I don't think reflects reality even as I see us beginning to see some trends back into the hospital.
• 14:15
  But the, we're never going back to the almost zero to, to 5% of adoption in the telemedicine space that we saw pre-pandemic. So, so that absolutely is here to stay, and there's a lot that AI can do in this space.
• 14:31
  The other piece, the sub-bullet around virtualization, is the expansion of the care team.
• 14:37
  And what I mean by that is how do things like chatbots or extended care team begin to do some of the tasks that only providers and specifically MDs did before?
• 14:50
  And I think we're going to see this more and more, and with the aid of artificial intelligence empowering folks across the care team, um, we're, we're going to see huge rises in this over the next five to 10 years.
• 15:06
  You know, I'm struck by the idea that, you know, in many ways COVID confirmed some of the thinking of folks who were on the edge of health innovation. This is the way we're h- we're going.
• 15:16
  And yet, ultimately there's, there's always things that really radically change our thinking. Yeah.
• 15:21
  And since you are dealing, you're in the environment of change management trying to, to push these things forward, I wonder how COVID changed your thinking.
• 15:30
  Didn't just, uh, a- affirm your thoughts about kinda where we're heading, but it could be your department or GE. I'm curious about kind of pivots that were made, uh, really kind of changes in thought. Yeah.
• 15:46
  Another good question. So, you know, it really killed anything that was technology for technology's sake. Mm.
• 15:55
  There was zero ability of our customers to focus beyond products that help them be more efficient and drive higher quality. Um, now, what we're seeing a little of that return, especially in high-end academics, right?
• 16:11
  There's always going to be this fabulous research frontier that's trying to push the boundaries of science. Mm. But I, I really think COVID, and, and look, in, in my AI world, brought
• 16:25
  in some ways backs back to the basics, right?
• 16:28
  Does this, and this is a little bit back to what I said at the beginning and really where I'm trying to keep the team focused is, look, if this AI product, uh, doesn't matter how beautiful your sensitivity and specificity is and the, and the, the elaborate designs of the underlying network architecture of your data science, if, if it doesn't allow a clinician to be more efficient, if it doesn't allow, um, a patient to have a better experience, if it doesn't raise the bar in quality, um, we're gonna struggle to see adoption.
• 16:59
  And so I, I think that's, um, a little- Mm... bit of kind of some of the refinement of thinking coming out of COVID. That's good. That's good.
• 17:07
  Um, I wanna sh- shift gears a little bit and get into this idea of, uh, collaboration.
• 17:13
  Um, big theme here at Startup Health, really our thesis is that you can't achieve health moonshots without really radical collaboration between startups, with industry, uh, with companies like GE.
• 17:26
  Um, I know you have made clinical partnerships a real keystone of your, um, just how you work. Uh, whether it's UCSF or Partners, uh, Healthcare or Boston Children's.
• 17:38
  So, um, I, I wonder what is your strategy around partnerships.
• 17:43
  We can sort of break it into two parts, sort of the big strategic partnership, uh, clinically, and then we can start talking about, you know, sort of the developer community and partnering with startups.Yeah, I, I think, I think, look, the, before diving in, and I usually think about three pillars for partnerships, the clinical technical partners and more of the, uh, the startup ecosystem.
• 18:07
  But b- before even diving into kind of more of the mechanics of that, um, the ceiling is so high in what we can do with data and artificial intelligence in healthcare.
• 18:21
  There is no one company that should be trying to crack this problem on their own.
• 18:26
  And, and just tiny data points, again, you guys live in this space, you'll know, but we've never been as good at a healthcare industry as creating data and as bad at using it, right?
• 18:36
  I mean, we, the, the, the stats that I see are anywhere between 2 to 4% of data gets used within healthcare to drive insights.
• 18:44
  The rest sits in some database in, in deep storage, um, in silos separated from each other, and just doesn't get used to drive anything actionable in healthcare. So you've got this incredible backlog of data. Mm-hmm.
• 19:00
  And as we actually interview providers, 80% of them are willing to adopt AI-embedded products, but only 10% have. So why? Why?
• 19:10
  You know, we haven't yet, as, uh, technology innovators and product innovators, been able to kind of cross that bridge, and I think a lot of that gets underpinned by building the right type of partnerships.
• 19:22
  So, uh, clinical technology and kind of the startup ecosystem. Clinical, um, Logan, you, you, you laid it out, um, well, uh, you know, we...
• 19:33
  [laughs] It starts by are you solving the right problem before you jump to the technology, right? Again, I,
• 19:41
  maybe this isn't what you expect to hear from the AI person, but the quicker I can get people to s- to, to think AI as a how, not a what, uh, the, the better.
• 19:53
  And so, you know, we've invested deeply at GE Healthcare in these clinical partnership on the AI front for two big reasons. One,
• 20:01
  uh, if you don't select the right use case, you mobilize a lot of brilliant people to build things that don't matter, right?
• 20:09
  So how, how do we ensure that when we mobilize our data scientists, when, when we mobilize our engineers, when we mobilize our, our commercial armies, it's all around a product that matters to our customers.
• 20:21
  We've got to walk in the shoes of our clinical partners and really, really understand their workflows. Um, so that's, that's number one. Um, n- number two is around technical partners.
• 20:31
  Um, this is a theme you, you'll probably hear f- from me kind of over this hour, but, um, I think one of the most important things in, in healthcare moonshots is, is knowing what you're great at and understanding where you need to partner.
• 20:44
  Um, at, and at GE Healthcare, we have been living deeply in technical workflows for decades, if not over a century, I think, with our x-ray products.
• 20:53
  And so we, we understand voice of customer and that last mile in the healthcare space of how do you deliver something.
• 21:01
  Um, we are never going to or want to compete with someone like an Amazon in the pure basic data science toolkit that they're building out.
• 21:11
  So four years ago, we created a partnership with Amazon, leveraging their SageMaker technologies, which are data science toolkits, and embedding that within our Edison platform for building AI, allowing us to focus on the data management and the data annotation capabilities that are healthcare and use case specific.
• 21:31
  Those pieces take an incredible amount of depth of knowledge in the healthcare space and an incredible amount of work, and partnering with the right technology folks allows you to move faster there.
• 21:41
  Similar partnerships with NVIDIA in, in the type of GPU work that they do better than anyone else, uh, with Intel, with some of the, the CPU work that they have embedded across our medical devices. And so I, I think
• 21:56
  thinking in that partnership first and where do I have a superpower and where do I need to partner with others to move faster is, is really pillar number two.
• 22:05
  And then pillar number three is around this, uh, in- incredibly fast-growing, innovative ecosystem, um, where w- we as GE need to move at the pace not of GE, but GE plus others.
• 22:21
  Um, and as we think about the Edison developer program that, that we've, um, been running for over a year and a half now, we really think about, um,
• 22:31
  carefully, are the voice of our customer, which the voice of our customer says, like, "Look, Carly, I want to tap into all this innovation, but I can't afford to have
• 22:39
  35 separate contracts, 35 separate billing experiences- Sure... 35 separate cybersecu- security checks, life cycle maintenance across all of these different solutions. Help, help centralize this for us." Sure.
• 22:51
  And for us, that means thinking about how we become better partners. Um, a- and that's a big focus, um, uh, that we, that, that we bring to life through our Edison developer program. Love it. That's great.
• 23:03
  Um, I love the insight that you, you see these partnerships as ways of getting really practical about problem-solving.
• 23:11
  It's not just about your sort of market opportunity, but, like, if you're not partnering with a, uh, a clinical entity, you're not able to walk in their shoes and talk to the person using the product.
• 23:23
  So that, using those partnerships as a way to get to the practical problems. Um, I also loved just this idea of GE plus others, creating that ecosystem.
• 23:34
  Um, to that end, I want to get to some questions in the chat and bring in some voices from the audience. So, uh, let's start with you, uh, Mehmet from Cleexa.
• 23:43
  If you can come off of, uh, mute, explain, uh, what you've built, and ask your question.Thanks, Logan, and nice to meet you, Carly.
• 23:52
  I think I was kind of impressed with your kind of definition of how you're doing this horizontal approach that focuses on imaging for sure. Um, we are a chronic care management patient engagement platform.
• 24:05
  Actually, one of them is oncology, and we have other platforms we built that we enable cross communication for patients and clinicians for multiple morbidities.
• 24:15
  Either, um, like, with their patient report outcomes is our focus, but, uh, one thing that resonated with me when you were saying standardization of the data for AI, and we're doing the same thing, how we can standardize information for both clinician and the patient to kind of make them clinically efficient and, you know, basic, the value-based outcome.
• 24:35
  So my question was really related into something that you said about, like, this horizontal view. We, as company, see that patient-reported outcomes is huge on any data that we collect, even with AI, right?
• 24:49
  So think about a patient that actually gets a- an oncology treatment. Your success is maybe rated with the cancer cells are getting smaller, but when you ask the patient after five chemo- Mm...
• 25:00
  sessions what their quality of life, that's a whole different story.
• 25:04
  So with that said, how do you incorporate with companies, even startups or other partnerships, that GE looks into the picture and say, "This is a good way to partner with, uh, a, a company to kind of, um, like, make a dent on this healthcare problem," like you just mentioned?
• 25:21
  Or you can tell us, like, some stories or some successes you did in the past and kind of give us a guidance on how we can partner.
• 25:30
  Oh, there's so much goodness in this question, so let me, let me try to unpack it a little bit. First, uh, Mehmet, you know, ha- happy to follow up offline if I don't touch on everything. But w- um,
• 25:41
  let me just give one stat before addressing it that struck me this weekend as I was, as I was reading. The... I really hope this transformation in virtual care stays, and I'm really nervous about it going away.
• 25:56
  One of the things that got me the most excited was when I saw a stat that 50 to 60% of patients involved in chronic care treatments are open to maintaining these virtual connects.
• 26:11
  Because we all know the, the volume in dollars [laughs] of healthcare really gets surrounded by this chronic care, um, uh, community and, and, and set of patients.
• 26:21
  So I, I, I, I love at a high level what you guys are looking to build. Um, I think patient-reported data inputs, um, are vastly reported, or vastly, uh,
• 26:40
  under-collected when you're building, building the patient jacket right now in patient view.
• 26:44
  And, and, you know, at, at a very simple level when we talk about the Edison platform, what we're trying to do is build ways to unlock data and then apply AI, right?
• 26:54
  And I, every customer I talk with, this unlocking and aggregating data across data silos, and even from not just brick-and-mortar healthcare, but also patient-reported or some of the, the more,
• 27:11
  the more wellness view, it's just we have a long way to go on that. Um, and so one of the ways we're addressing it at GE, and even thinking about partnerships, is, um,
• 27:24
  trying to, l- look, trying to move beyond our own P&L structure. And so, uh, you know, look, that, that's one of the challenges of any large company.
• 27:32
  And so specifically we've tackled oncology and cardiology as care area focuses, where the, the go-to-market motion and the product development roadmap needs to be focused on those care areas first, and the product segment second.
• 27:51
  And so what that means is we're very aggressively in those two areas, looking to bring the data together and bring the partnership ecosystem together to build out the spaces that, that we're not going to do natively on our own.
• 28:06
  And so I, I think having, um, a validated use case that people want and people will pay for is the most important thing for driving a partnership conversation.
• 28:20
  Um, and, and if you can bring those things to the table, um, and show how we can go, go to market stronger together, it's the perfect opening for a partnership conversation. That's a great question.
• 28:36
  Thanks for the question, Mehmet. Uh, let's, uh, go to another question in the chat. Uh, Natalie Davis has a couple of sort of, um, nuts and bolts questions about how GE partners with startups. Um, go ahead, Natalie.
• 28:49
  Thanks, Logan. Um, hey there. I'm Dr. Davis. I'm a pediatrician and chief medical officer of PreventScript.
• 28:57
  At PreventScript, we partner with primary care providers in the outpatient space, upstream, before cardiology and oncology ever happens, uh, to help their, um, help them identify and intervene with their patients that are at risk for diabetes and hypertension.
• 29:15
  So really targeting that upstream, overweight, pre-diabetic, pre-hypertensive, before they get cancer group.
• 29:23
  And, um, we pair that with a, a series of interventions that are, uh, remote monitoring and clinic, clinic-based. So asynchronous telemed if, if, if you were, if it were. Yep.
• 29:36
  Um, and so I guess my questions are just sort of nuts and bolts questions around, um, your stage of partnership. Um, do you lead investments? Do you co-invest with groups like StartUp Health?
• 29:50
  And, um-Average check size Thanks And smallest check size. Thanks, Natalie. Uh, I love this one. So first of all, Natalie, my, my husband is a primary care doctor at Kaiser. All right.
• 30:03
  And so keep doing what you're doing because, uh- Thank you... you know, it'll, it, it's, it's so, uh, it's so important, and this was actually from my time at Apple.
• 30:13
  Being able to quantify the impact of this upstream work is, is so challenging to get right. We all know it's the right thing to do. Yes.
• 30:22
  But how do you get people to pay for the right thing to do and, and drive that quantification? So I, I love- Welcome to my world. I, uh, I lo- I love it. Um, so, uh, you know, um, we do not do...
• 30:36
  GE used to have a venture arm. Um, that's not something that we focus on anymore, and so in our partnership space, um, it's not direct invest.
• 30:46
  We do indirect invest in terms of, you know, uh, te- technology integration support, um, go to market support, marketing support, that type of thing. Okay.
• 30:55
  Um, we, we also run regional acceleration programs where we do dedicated, uh, kind of mentoring, but we do not right now have, um, a vehicle for doing direct investments- Okay... um, in, in early stage companies.
• 31:09
  So it's more finding people right now where the strategic alignment exists- Mm-hmm... and then putting our money where our mouth is with our time and resources, but not, not, not,
• 31:20
  not, not a check as, as an investment partner. Right, right. Reseller? Reseller agreements, though, with your existing customers? Yep.
• 31:29
  We're, we have a marketplace right now with over 100, uh, solutions listed across GE products and third party part, products.
• 31:37
  Um, we tend to believe that the value prop gets stronger to our customers if there can be some integration and one plus one equals three- Yeah...
• 31:48
  um, a- as opposed to just a pass-through opportunity, but we do, um, have some reseller opportunities, uh, through our marketplace. Right. Very good. We'd like to learn more about that. Thank you. Of course. Yeah.
• 31:58
  And we'll be interested to see what your husband thinks of our demo. [laughs] He might like it. All right. Awesome. Sounds good. [laughs] Sounds good. Th- thanks, Natalie.
• 32:06
  Uh, let's go to, uh, Dorothy Goldman from Oritel Diagnostics, who's got an interesting question in the chat about sharing, uh, images a- across institutions or from the home, uh, into the hospital.
• 32:18
  So Do- Dorothy, if you could come off mute. Okay. Hi, uh, Carly. O- Oritel Diagnostics works with, uh, imaging in a different way. We use colors and we test, um, saliva for d- di- diagnosing endometriosis.
• 32:32
  And I see a challenge there where you can get a image result and you get data, but you wanna sh- it's d- it's useless to the patient unless they have direct access to a clinician, but then the clinician a- it also needs to be reimbursed with insurance and other agencies.
• 32:50
  And I just wanna know what your experience is in the pathway of what happens to this data once you get it, and how do you make use of it, and how does it fit into the total big healthcare p- picture? Yeah.
• 33:03
  Great, [laughs] great question. So j- uh, as, as you're implying with, with, with your question, just because we create data doesn't mean we have access to use it for AI by any means, right? Yeah.
• 33:15
  The data is owned by the institution, and to some extent, depending on the country, um, the patient as well. Yes. And so- Complex. [laughs] It's, it's very complex.
• 33:25
  And so this is part of the reason, um, wh- uh, why we had to invest in platforming early on, is because we couldn't leave it up to i- any of our individual product teams to follow the rules and regulations and legal guidelines of every single data sharing opportunity.
• 33:44
  We had to build a playbook around that. Um, and so, you know, we work, um, in building out a base, if you will, of data partners at the provider level across different
• 34:00
  geographic, uh, institutions. And then we almost set up an opportunity for multiple people to use that data for various AI use cases or data needs that they have. Um, and, and, and, and
• 34:17
  I could probably talk a little bit more offline. Some of this is I've, I'm being a little cagey about because we haven't totally been public around some of, some of this data work.
• 34:25
  Um, but right now we work very much at the provider level, not the patient level, just for, um, keeping a focus on scale and speed in terms of building out our, our data agreements and, and data access agreements.
• 34:39
  Um, did that get at your question or, or re- Yeah, yeah... reframe it back to me? No. I, I think that the regulatory issues too are also in the development process,
• 34:47
  because I know a few years ago there wasn't any regulation. [laughs] Sure. And now it's very urgent that there is regulation, and just understanding how that works is complex. Yeah.
• 34:58
  And, and I've, I've, I've spent some time with the FTA, and, and one of the things we're, we're trying to teach them when it comes to AI is, uh, um, volume alone isn't sufficient. So we, we call it the three Vs. Mm-hmm.
• 35:13
  Like, when you're building data science and AI solutions, there's a lot of talk about, like, volume, volume, volume. But if you get a million images from one city, there's no way that's gonna generalize across a globe.
• 35:25
  And so volume, variety, meaning you have to have data that represents the populations that you're solving for. So volume alone isn't sufficient. You've gotta get variety.
• 35:35
  And then this veracity pillar is around how you annotate it, right? Garbage in, garbage out when it comes to this AI training.
• 35:41
  You can train on millions of images from broadly around the globe, but if you've circled cats and called them dogs every time, you're gonna have a bad AI model.
• 35:50
  And so we, we are trying to help-Um, regulatory bodies think this through as well.
• 35:56
  Um, be- and, and to their credit, they saw this coming three years ago with this tidal wave of, uh, applications and, and, um, you know, 510[k]s to PMAs coming across their desk and them not having the staffing to keep up.
• 36:13
  So they've been trying to change their frameworks at the right speed, but it's... It is one of the challenges of healthcare, um, that that regulatory guidance often lags the technology and product capabilities. Yeah.
• 36:28
  Yeah. Awesome. I appreciate the question, uh, Dorothy. Um, I've got a que- uh, it looks like Moira Scheck from QBISME's got a question. Um, Moira, can you come off of mute and go ahead and ask it? Hey, Logan. Thank you.
• 36:42
  I'll put on my, um... In transit across my house. Um, nice to meet you. Um, yeah, so we have... QBISME is a, a cloud-based solution. I'm a clinical radiologist and a cancer imaging expert.
• 36:53
  Um, and so, you know, from my perspective, you know, trying to, you know, get a huge transformation that we need, what we're doing today, you know, belongs in the ice ages, um, for cancer imaging clinically.
• 37:05
  Um, so, you know, from my perspective, we really, um, design so- that, um,
• 37:13
  you know, we feel is most powerful for, um, you know, revolutionizing how a radiologist with interfaces can interact with data as an augmented intelligence solution, um, for cancer, um, and for, you know, general practice as well.
• 37:29
  Um, and what we have on the back end is, um, a patented technology.
• 37:32
  I've held the patents for five years using dense stitching of a variety of data, like multiple types of MR and dense classifications, um, and, uh, uh, prediction stitching, which is the critical need technology that we need, um, for cancer, and this needs to be cloud-based, right?
• 37:52
  Mm-hmm. Um, so, so when we're thinking about, um, you know, what GE's doing, um, and, you know, and I, I don't wanna embarrass you, but these are really important conversations.
• 38:04
  Um, and what we found is there's a lot of dominance, um, in radiology by a lot of, um, groups who, um, have worked with machine manufacturers for many years, you know?
• 38:14
  And there is, like you said, a huge push to put the deep learning on the machines, and I do think that there are some great use cases where you can do that.
• 38:23
  Um, but, you know, there's a really central and important ethics issue here.
• 38:28
  Um, and there was a great lecture by a top Oxford ethicist at AI Med recently, um, and how he described it, first of all, he said, you know, this isn't artificial intelligence, you know?
• 38:39
  And, and, um, these are artifacts. These are computer artifacts. You know, and, uh... or the term artificial intelligence, I believe, was really pushed by industry, um, who really wanted to hype this up.
• 38:51
  And to lay people, they don't really understand it's just a powerful classifier, right? So that's number one. These are just artifacts.
• 38:58
  And when you have an artifact making a decision, so all of a sudden we have an artifact on a machine making a decision, then you're, you're...
• 39:07
  All of a sudden, we have a new huge issue here, where we're splitting responsibility away from the doctor, right? So, so to me, um, you know, I, I had a demo. You know, there's...
• 39:21
  So there's the ethics argument, and then for me as a clinical radiologist, there's also the, kind of the, the practical side.
• 39:27
  Someone did a demo because they had triaged where a machine had used an AI algorithm, which by the way, the AI algorithm's on single images, which are...
• 39:35
  It was very different from what QBISME is doing, which is dense like, just like you said, we need dense stitching of many types of data that's much more powerful, and it's gonna be a lot more accurate.
• 39:44
  So these algorithms running on single images using neural networks, um, 93% of radiologists say they're unreliable in clinical practice, right?
• 39:54
  And I had a demo where somebody was doing an automatic triage of a chest X-ray with a chest tube, um, and calling it urgent.
• 40:03
  And actually, when they showed it to me, this was really embarrassing as a clinical radiologist, it had a chest tube in place, which means- Mm-hmm... that the black box was, it's brittle, it's not intelligent. Mm-hmm.
• 40:14
  Couldn't figure out and didn't have the common sense to know that's no longer a, um- Yeah... an emergency, right? So- Yeah... so these are really, really complex ethical issues.
• 40:26
  And when you're splitting a decision-making away from a doctor, that has to be done with enormous care. So, you know, so I'd love to hear you just talk about, you know, how you view that, um- Good question...
• 40:39
  and how you guys are gonna approach this, um, so that, you know, that, that, you know, you're thinking about all the potential risks of splitting that responsibility away from the doctor. 'Cause I strongly...
• 40:49
  It's my strong opinion, um, that that shouldn't happen, that it- Good... needs to be augmented intelligence, that it's augmenting the doctor.
• 40:56
  Um, but I do think there's some great use cases, like for example, anatomy labeling, so we can reorganize data, is an awesome use case, um- Great question, Moira... for, for GE to, to pursue. Great question, Moira. Okay.
• 41:09
  Yeah. So, so Moira, uh, uh, you may be surprised to hear me say I 100% agree with everything you just said. So, so let me, let me give a little bit of framing to even how we approach AI here at GE Healthcare.
• 41:22
  The grand, grand, grand majority of our organically developed use cases are what I call upstream versus downstream, and this is, this is [laughs] different than what we were talking about in the primary care space.
• 41:33
  But upstream, meaning how do you use AI to create the data faster, cheaper, potentially lower dose, those kind of things.
• 41:43
  So an example is, uh, an MR use case where using deep learning in the reconstruction phase, we were able to create images with higher quality at less time. You'll know this, right?
• 41:55
  That the age-old p- uh, paradigm in MR has beenTime equals quality. And so what we're able to do is apply artificial intelligence into the imaging chain, and to create better data, faster data, cheaper data.
• 42:12
  When it comes to downstream and actually driving clinical decision-making,
• 42:18
  we're really not playing in that space because we think it's going to be the ecosystem of deeply specific startups and innovators who go deep in a specific area who can solve those problems.
• 42:34
  And to be honest, what's gonna happen first is everything you just talked about, which are much more about, um, time savings, not clinical decision-making.
• 42:45
  I mean, to the FDA's credit, they've held that bar really, really high on clinical decision-making.
• 42:50
  And neither you or I wanna see what happened in the CAD space happen again with radiology, where every, uh, every, every breast imager I spoke to said, "Look, I... We use CAD because I get reimbursed higher.
• 43:02
  It doesn't actually help me do my job better." Right? That would be, that would be the disaster in this space.
• 43:07
  Um, and so I, I think our strategy is pretty well aligned with what you've just said because where, where we focus is AI to make data better.
• 43:20
  And what we're trying to set up in terms of the ecosystem approach is, again, to what you said, really investing in orchestration capability.
• 43:30
  So the ability to have a platform, tool, and structure, you can think roads and bridges, that sits across many, many, many different decision-making or not even decision-making, but sometimes more operational focused AI downstream solutions, but tease them up in a way and delivers them in a way that's unified, um, and more structured back to the radiologist, right?
• 43:56
  Because the, the other thing I hear constantly, and, you know, love your take on this too, is radiologists don't want 1,000 pop-up screens within their cockpit, right?
• 44:06
  You've already got four panels of screens that you're looking at between work list and, and comparing priors.
• 44:13
  And so part of what we're trying to do is kind of create a technology layer that aggregates across multiple different solutions, and then teeing it up for the final decision-making to still be done by the clinician.
• 44:24
  I think we're very, very far away, um, if we will ever get there, from, uh, you know, the, the clinician not having to be involved in that final decision-making.
• 44:36
  I, I really see the next 10 years of AI and radiology, um, being clinician plus AI. Um, now hopefully AI makes things more efficient.
• 44:45
  Hopefully AI can do some of the manual quantification that isn't a good use of doctors' times. Um, but the clinical decision support, that's not something that can or should move away from MDs anytime soon.
• 45:00
  I don't know if that addresses the question. I think it's a, a wonderful answer. Please. I, I love your answer, and I think, um, what you just said is that you're gonna use your technology to help organize the data.
• 45:11
  Um, and that's a perfect answer. And, you know, for me, if I'm looking at a brain MR, I can be looking at a normal brain MR, and I'm looking at 22 sequences. You know? Yeah.
• 45:19
  Because somebody's got 22 sequences on a normal brain, um, and, you know, so I'm spending, you know, an enormous amount of time looking at normal 22 times. You know? Mm-hmm. Awesome.
• 45:29
  So, so the whole, um, joint effort of getting all the data organized around anatomy, which is what we're doing, um, is, is a fantastic goal. So thank you. Thanks, Moira. No, great, great question.
• 45:41
  Thanks for the, thanks for the answer, Carly. I wanna shift gears. We're getting towards the top of the hour already. Uh, time moves fast, and I wanna get some of your really practical wisdom about, uh, startup teams.
• 45:53
  Um, you've assessed a lot of solutions. You've worked with some top tech teams in the industry, uh, and you've seen what makes some work and, uh, some not.
• 46:02
  So, um, I wanna get your advice for the, you know, dozens of founders on the call about what you think the key ingredients are for a, a successful health innovation, uh, team in particular. Yeah.
• 46:17
  Um, look, I'll start high level, and don't roll your eyes at me, folks. But if you don't have the right purpose and the right people, none of what I say after this matters. Um, and that's not healthcare specific.
• 46:28
  That is any company I've ever worked in.
• 46:30
  If you can't articulate the why, pe- talented people can go work anywhere, so give them a reason to choose what you're doing, um, and surround them with brilliant people who can, uh, challenge them, help them be better at what they're doing.
• 46:45
  And so look, if you don't have the people and the purpose, i- nothing else matters. But I do think there's three other ingredients when it comes to digital health that you've gotta get right. First, and, uh,
• 46:59
  do you understand the product fit, [laughs] right? Will, will people actually... Do people ac- actually want this problem to be solved? But then just as importantly, will they pay for it to be solved?
• 47:12
  So the product fits and the commercial model, and I often see companies get one right and the other wrong, and then you stall out. But then the third piece that you have to think about from day one is regulatory, right?
• 47:25
  If you think about it in like a hierarchy of needs, like product, will anybody use it? Commercial and business model, will anyone pay for it? Regulatory, will it ever get to market, right?
• 47:36
  And y- you have to have all three of those pieces together in the digital health space, um, and, and from pretty early, almost like the first three hires, um, within your organization, um,
• 47:48
  uh, if, if you wanna get it right.
• 47:49
  Now, I, I think one thing I'd be, um, amiss not to mention is this, this balance of healthcare domain versus outside in thinking, another thing essential to get right, and there is not a perfect formula.
• 48:04
  I think we've all watched-Tech first companies struggle to get into healthcare.
• 48:09
  Um, but that doesn't mean you should have a team of only healthcare domain startups and experts without some people to challenge the, "Well, this is how we've always done it," thinking.
• 48:22
  But you do need to get that balance right i- if you're gonna be successful in digital health.
• 48:29
  I, what do you think, uh, is the, uh, how do you recommend for startups who think that this idea of resiliency, and we, we just came through this year, uh, of turbulence and challenge,
• 48:41
  uh, a- and I wonder what is your advice for sort of weathering some of those storms? Uh, yeah. I, I mean, this is one of those life questions you gotta figure out, like, h- h- how you get it right.
• 48:54
  But I guess for, for me, there are two things I thought of here. So one, again, why are you doing what you're doing? If you have the right why in place, you're, you can weather a lot of storms.
• 49:07
  And then two, this concept of personal and professional board of directors. I think I'll, I'll, I'll speak personally in my own life.
• 49:15
  Um, resiliency really depends on the community you have around you and your ability to have people hold you accountable, but also trusted people where you can be vulnerable with things you're thinking about, challenges you're having, and you can trust that the advice that comes back is purely to help you be, to help you improve.
• 49:34
  And I, I, I specifically say professional and personal 'cause I, I, I think you kinda wanna have those two groups, um, if you wanna set yourself up for resiliency in, in a space where it's hard to make the changes we all wanna see in terms of driving some of these, these technology and step one changes in healthcare forward.
• 49:56
  Where do you see specific opportunities in, in collaboration with, uh, with startups? You've mentioned, uh, imaging. We kind of went down a couple different specific roads.
• 50:07
  Um, you know, Startup Health sort of represents a broad global ecosystem of solutions and problem solvers.
• 50:15
  So sort of outside the norm, where are you seeing some interesting opportunities for, for innovation and collaboration within your world?
• 50:24
  Yeah, you know, the, the clinical spaces where we focus most specifically, radiology, cardiology, oncology, anything in that space, but then the broader trend of virtualization, w- we aren't going to do that
• 50:37
  on our own as a company that has for decades worked with providing devices to brick-and-mortar hospital settings.
• 50:46
  But virtualization is going to be, in my mind, the number one trend that flips healthcare on its head in the next decade, and there's sub-trends there with moving to cloud and, and, and AI and interoperability.
• 50:58
  But I think the, the big trend of virtualization is an area where we are looking to partner to move faster than what we would ever do to build something on our own. Um, we're getting towards the top of the hour.
• 51:14
  Something we do every Fireside Chat is we give the folks on the call a chance to, uh, reflect back to you and to the others on the call their greatest insight from the call, uh, something that they've heard, uh, that they want to, to sort of reinforce and share.
• 51:28
  So if anyone would like to, to share their insights, you can drop your name, uh, or your thought into the chat. Um, and, you know, I'll start.
• 51:40
  I think what, one insight that I'm still really thinking about is the importance, uh, of these clinical partnerships.
• 51:48
  And, uh, oftentimes I think a startup can think about a, a hospital partner or hospital system partner as a, uh, a foot into the market, as a market expansion opportunity.
• 52:00
  And you talked about the importance of clinical partners as a way of just understanding the real problem.
• 52:07
  Like, if you're not solving a real problem for somebody, some- you know, uh, if you don't, can't get into their shoes and see that problem, then you're not gonna get anywhere.
• 52:18
  Um, and so that really, that, that thought really struck me. Um, Natalie, you've got an insight. Go ahead and share.
• 52:31
  I just thought it was great to hear you say, uh, Carly, that, uh, you understand that you're inside the hospital in cardiology, radiology, oncology, all the, we call them ologies, and, um, yet you understand how big virtualization is and is going to be, and th- that you want to partner on, to that end.
• 52:56
  We absolutely do. I meant every word. So if any of you wanna talk more, let me know. [laughs] Uh, w- we've got time for another, uh, if anyone would like to share. I know, uh,
• 53:10
  Joel from Keep, we didn't get a chance to go through your whole question, uh, but we might have a second if you'd like to share your thought or an insight. Oh, it's good.
• 53:22
  I [clears throat] it's a company that's moving towards using machine learning and, and AI.
• 53:26
  It was, it was mostly a question of the difference between your roles and how you've implemented it in the past at a company that is more positioned itself to dedicate resources to, uh, research based programs to GE, where you have several different pillars of PNL.
• 53:41
  And, um, so I, I, I got a good idea of, of your vision for, um, moving forward, but I'm wondering how you choose. How do you choose what products to, to attach it to? Uh, Joel, it's an awesome question.
• 53:54
  So, you know, in the very, very early, I, I, my organization now is about 250 people, uh, you know, 60 data scientists, 120 so engineers. So we, we have a lot of-... scale that we've built to over four and a half years.
• 54:10
  But four years ago, it was three data scientists and seven engineers, and we more or less, I, I mean, to be frank, ran reverse shark tanks within GE HealthCare around, like, around which use cases would drive the most movement for our customers and their patients, and most movement in the market, and then we prioritized those first.
• 54:34
  And we got wins on the board that allowed us to scale what we were doing as an organization to then spread to that next level of use cases.
• 54:42
  Um, a- a- and so, you know, that's, that's a little bit how we built the model, um, at GE HealthCare in terms of, look, it's...
• 54:52
  At a, at a large matrixed organization, you've got to be able to prove that you can defend the core while you build the platform capab- uh, capabilities to move into one step adjacent opportunities.
• 55:04
  Um, and that was really how we set up the organization from, from the beginning at GE HealthCare. That's great.
• 55:10
  I mean, I think my biggest insight then is even at the huge industry level, customer still comes first, and, and that's a really valuable thing to, to remember, so thanks. Thanks, Joel.
• 55:21
  Great la- great last question, uh, Joel. That takes us to the top of the hour.
• 55:26
  Uh, Carly, thank you so much on behalf of everyone on the call, uh, and everyone who will, you know, view this in the future, uh, thank you so much for taking the time.
• 55:34
  I, I, I can speak for probably others on the call when I say, uh, I'm more excited about what GE HealthCare is doing, uh, knowing that you are involved, and hearing your sort of moonshot vision for partnerships and for, uh, the ethical framework behind AI.
• 55:52
  Um, so really appreciate you, uh, sharing those thoughts and just hearing your, your vision for the future. My pleasure, and thanks for giving me the, the chance to connect with all of you guys, so thank you.
• 56:04
  [outro music] StartUp Health invests in health transformers from around the world who are committed to achieving audacious health moonshots.
• 56:19
  If you wanna learn how you can join this community of entrepreneurs, or if you wanna connect with one of our 330 companies, go to startuphealth.com. Thanks for listening to StartUp Health NOW. We'll be back next week.
• 56:32
  [outro music]