73 – AI and Healthcare Innovation: A Conversation with Dr. Rick Abramson

Marcus: [00:00:00] If you enjoy this content, please take a moment to rate and review it. Your feedback will greatly impact our ability to reach more people. Thank you.

Vic: Rick Abramson, welcome to Health Further. Excited to have you on. We have been digging into AI related to healthcare and generally following AI probably for a year, year and a half when it first started coming out and excited to talk through this with you because you’ve really been at the, at the forefront of AI even before any of us were paying attention.

Vic: And so maybe let’s start off by giving a little bit of your background. You’ve had a pretty diverse career. You’re an MD, Harvard, uh, medical school, but then you’ve had a, a quite diverse career in different areas, consulting, non profit, for profit, startups. Give the audience a little bit about your background.

Rick Abramson: Yeah, sure thing. Um, thanks, uh, for having me, Vic and Marcus, um, been looking forward to, [00:01:00] to this for a long time. Um, uh, so, um, let me just start off by saying that I, I consider myself a healthcare generalist and I say that because, uh, over the years I’ve acquired a pretty, uh, Pretty deep, but pretty niche experience in this whole area of AI, uh, in medical imaging.

Rick Abramson: Um, but you know, I’m here out of the conviction that, um, uh, you know, despite the kind of niche, uh, niche aspect of this expertise, I think there’s lessons to be learned, um, that can apply across the whole healthcare ecosystem. And, you know, obviously there’s a, you know, the listeners to this pod, uh, kind of span that, that ecosystem.

Rick Abramson: And so, um, you know, over the conversation, I think, you know, let’s try to, to draw out those lessons, um, you know, as much as we can. Um, so as far as my background goes, um, you know, I, I have kind of, uh, this, uh, kind of checkered checkered past as, uh, you know, kind of a clinician and a non-clinician kind of hopping back and forth, uh, you know, over the divide between the clinical and, and nonclinical, uh, parts of the healthcare realm.

Rick Abramson: Um, trained as a, as a physician in particular, a radi. [00:02:00] So I’m a board certified radiologist and spent about 15 years in clinical practice, including 10 years at Vanderbilt on, on faculty here in, in Nashville. Um, but to your point, also had a lot of experiences over on the, on the nonclinical side. I actually started my career in the policy and health services research area.

Rick Abramson: Um, did a stint at the white house doing Medicare analysis, um, put in my time with McKinsey. Uh, in company as a consultant, uh, and then later on, um, here in Nashville, uh, served a couple of years in an executive role at HCA, uh, before jumping over into this, uh, AI world, uh, where I’ve been for the past three or four years.

Rick Abramson: Um, and since I think we’re probably gonna, you know, focus a lot on, uh, on AI, let me just tell you real quick about kind of what that that last jump was like and what motivated it. Um, so, uh, sitting over at HCA, um, doing this physician executive, uh, role, kind of have a bird’s eye view of, of, uh, what’s going on across the landscape across of [00:03:00] all, all the HCA hospitals.

Rick Abramson: And, uh, I’m recognizing there’s, uh, there’s some problems emerging. Um, I’m seeing procedure volumes going through the roof. I’m seeing the workload increasing on providers. Um, in the, in the meantime, I’m also seeing the supply of providers, the physician supply, uh, remaining, uh, the same or even going down, which means the workload on any individual provider is, um, increasing astronomically.

Rick Abramson: Uh, and what happens? Well, physician morale goes down, becomes harder to maintain quality, and we need a solution. And so, you know, like others, I started looking around and looking to technology as potentially, you know, providing the, the solution to, to all those problems at the same time. And that was really kind of the genesis for my, uh, really kind of wanting to do a deep dive and a real immersion into the AI world.

Rick Abramson: Um, and that’s kind of what led to my, my jumping ship and kind of, uh, joining the AI startup revolution over the past 34 years, um, first with a company called [00:04:00] Cobra, um, outside of New York that is, um, using AI working with payers and then subsequently with a company called Annalise AI, uh, which is out of Sydney, Australia, um, that’s, uh, developing advanced AI solutions for medical imaging.

Rick Abramson: So, um, you know, that’s kind of like the. Uh, that’s kind of the foundation, um, and, uh, learned a lot and, and, uh, you know, had a, uh, been able to kind of look at this, uh, this area from a bunch of, uh, different vantage points and, uh, you know, looking forward to kind of getting into it, um, as we, uh, as we talk.

Vic: Yeah. I want to dive into radiology as a specialty because I think it’s, it’s pretty interesting. Radiology was one of the earliest adopters of technology all the way back to x rays, but then more recently, uh, MRI and PET scans, CT, And then, uh, even with workforce, uh, shifting, with the, the night reads across the other side of the globe, uh, is that a, a proper frame of reference where radiology [00:05:00] has really been an early adopter of technology and sort of workforce optimization?

Vic: Or how do you think about that? Yeah, I mean, I think it’s a great way to

Rick Abramson: frame it. And, and look, I, I look at, at, uh, at radiology and in some sense it’s the, it’s kind of a bellwether. Yeah. Uh, bellwether field kind of leading the way and in some ways kind of forecasting what’s going to happen in in other clinical domains.

Rick Abramson: Um, you know, to your point, you know, this is kind of, uh, the story of of radiology has been one of, uh, progressive evolution and very quick technological evolution over the past 120 years, you know, starting in 1895 when this German scientist, Uh, Wilhelm Röntgen was playing around in his laboratory and found out that, uh, there was a piece of barium coated, uh, uh, barium coated screen that was fluorescing on one side of his laboratory and, uh, decided that, oh, hey, that’s, there must be some mysterious rays emanating from this cathode ray tube that I’m playing around with.

Rick Abramson: And, you know, [00:06:00] since he didn’t know what they were, what they were, what they were up to, he would call them X rays. And, uh, you know, that’s. That’s kind of how the profession got it started. All of a sudden, you know, we have this whole new medical field that progressed very, very rapidly. You know, x rays and then fluoroscopy and then ultrasound and MRI and CT and nuclear medicine.

Rick Abramson: And, um, you know, the, the history of this field, it’s been very interesting. If you kind of look back, it’s been one of, um, uh, successive, uh, uh, technological innovation, kind of game changing innovation every 10 or 15 years. Uh, kind of a new way of imaging the human body. Um, until, until about 1980, 1990 and around 1980 and 90.

Rick Abramson: That sequence continued. It was still a big, big, you know, shift every, you know, 10 or 15 years. But this time, instead of a shift in the way we image the human body, the shift started becoming the way we manage information and the way we archive and store information. And so eighties and nineties, you saw [00:07:00] the advent of this thing called packs Stanford picture archiving communication system is these big.

Rick Abramson: Consoles that radiologists now stare at all day in a dark room, uh, that replaced film and that was a huge revolution. If you think about it, I mean, like the entire, uh, film industry disappeared over the matter of a matter of like a year or two, all of a sudden, instead of printing out x rays on film, you’re looking at him on a, on a computer monitor.

Rick Abramson: And that’s the way it’s been, um, 40 years. And then 10, 15 years later, all of a sudden you’ve got this new advent called tele radiology, where you can transmit images, uh, for remote interpretations somewhere else, other than when they were acquired. Yeah, this is something that would not have been possible before high bandwidth internet.

Rick Abramson: And again, sudden. Sudden transformative change in how the field is practiced. That was kind of in the mid 2000s. And now I would argue we’re in the middle of the next incremental wave of transformation. So, so I think that the next, you know, 10, 15 years have passed since teller radiology. And now [00:08:00] we’re, we’re firmly in the, in the revolution of AI and we’re, we’re kind of, you know, building the airplane as we fly it.

Rick Abramson: But I think it’s dramatically beginning to already beginning to dramatically change the field. And I think it will continue to do so over the next couple of years.

Marcus: Can I ask a quick question before we continue on? Rick, what do you think is special about radiology where the clinicians are embracing technological advancement at the rate of every 10 to 15 years?

Marcus: And maybe now, you know, it’s getting into less than 10 years. You know, that’s that’s something that we always talk about as investors. You know, the rate of change that clinicians are willing to adopt. And, um, it seems like in radiology, it’s it’s advanced beyond what other specialties or primary care is willing to accept in terms of the step changes.

Marcus: So is there anything about radiology just from a [00:09:00] mindset about the practice itself that leads to a different sort of culture there in terms of the willingness to embrace change?

Rick Abramson: Yeah, it’s a really interesting question, uh, Marcus. And I think, I think it’s multifactorial. I think there is, um, there is a cultural aspect there.

Rick Abramson: I think that for whatever reason, whether Whether the, the field, uh, creates, um, uh, physicians who are more amenable to change or whether the physicians that happen to go into radiology are more innovative or more progressive or more change minded. I’m not sure, but there is that, that, that cultural element.

Rick Abramson: And, and look, radiologists, if you kind of survey radiologists, they’re, they’re very proud of that. They’re, they’re kind of very proud of being at the vanguard of technological change. It’s kind of like a, a core identity among radiology. Um, but I think there’s other aspects to, um, a lot of research funding, uh, over the years has gone into, uh, medical imaging and, um, and some of it is derived from other, I mean, like, I mean, there’s, you know, uh, research into, um, uh, like military spending in, in World War II eventually kind of found its way into [00:10:00] medical imaging applications, right?

Rick Abramson: And so, I think radiology has been the beneficiary of a lot of, um, of, uh, scientific research advancement. Um, and then I think the final thing, honestly, is just economic and, um, and, and kind of in competition and entrepreneurship. And I think the recognition among radiologists that if they don’t stay on the forefront of technological advancement, they’re going to get left behind.

Rick Abramson: Um, and, uh, you know, one thing about medical imaging is it is always embroiled in turf wars. Uh, with other specialties, um, and so, you know, radiologists, you know, the clinicians, kind of the clinician listeners, I think we’ll know what I’m talking about. You know, the radiologists are always fighting it out with cardiology and with OBGYN, with vascular surgery, and there’s always, you know, everybody’s kind of, you know, kind of fighting it out for the, for the more lucrative procedures and for the volumes.

Rick Abramson: Um, and I think radiologists know that if they’re not constantly advancing the field, um, they’re, they’re going to be obsolete.

Vic: Yeah. So, I, I’m interested building on that. Radiologists. I think have a really interesting role in, [00:11:00] uh, kind of interpreting the images for the other doc who’s delivering care. And there’s something around that partnership, uh, you, the radiologist needs to keep extending, keep getting more and more at the cost, at the frontier of new technology tools to advise the cardiologist or an internal surgeon or whatever the other practices are, orthopedic.

Vic: Thank you. Uh, they could do it themselves, but they would probably miss some things on the images. But if it’s a very standard image X-ray, maybe they can look at that. Uh, maybe a more complicated image. They need the, the true expert, the radiologist, to help, help translate. Is that fair or do you think that’s a mischaracterization?

Rick Abramson: Uh, I, I think it’s fair. I think it’s also something that’s changing, and I think this is actually a conversation that’s got, that’s really gonna come into the forefront with. The AI revolution, right? Because now we’re talking about software applications that could potentially read the image for you, or if not completely read the image for you, then [00:12:00] certainly help you read the image and help you avoid errors and mistakes and potentially expedite the, uh, the, you know, the, uh, the speed with which you come up with diagnoses.

Rick Abramson: And so there’s a lot of talk about using AI technology to empower. non radiologist physicians, and for that matter, even just non physicians, uh, to, to render diagnoses. And so, uh, you know, whereas once before it was, it was all the radiologists, uh, purview, you know, now you’re talking about, you know, using AI software to empower, um, ED physicians or ICU physicians, or potentially to empower Uh, you know, physician assistants or nurse practitioners, um, especially in underserved areas of the globe, uh, that may suffer from, from a physician shortage.

Rick Abramson: And so I think there’s going to be a lot of disruption and a lot of conversations about professional identity, uh, coming up in the future. And again, this is one of these areas That’s not specific to imaging. I think all throughout all throughout medicine, there’s going to be a redefinition of kind of traditional physician and provider roles, and it’s all enabled by the technology and it’s [00:13:00] all in service of this idea that, you know, you can, you know, you can automate certain functions and then potentially, you know, standardized and replace, you know, certain functions, uh, potentially with, uh, you know, lower cost labor input if you want to be very dry and economic about it, um, but it’s also a matter of trying to augment quality as well.

Rick Abramson: Um, through standardization and efficiency, and so everything is kind of, I think, on the table right now as we kind of sort through this whole, uh, emerging software revolution.

Vic: Yeah. Excellent. Uh, well, so before we dig into the AI specifically, I’d like to get your perspective on a large academic medical center like Vanderbilt.

Vic: versus a large for profit health system, HCA, versus a U. S. VC backed company versus a global company headquartered in Australia. Those are the last four jobs you have had, I think. How, how do those different, uh, enterprises, think about new technology in general and how do they function? What’s the [00:14:00] culture differences?

Vic: Can you compare or contrast any of those?

Rick Abramson: Oh, man, it’s super interesting. Um, probably deserves a podcast of its own. Um, uh, you know, I, I learned, I learned something from every, every stop I make. Uh, yeah, which is what makes it so, so interesting. Um, Uh, you know the I mean, I’ll take a shot at it from a very kind of high level.

Rick Abramson: Um, you know, the academic centers that I’ve been at and it’s not just Vanderbilt. I’ve been at Johns Hopkins. I’ve been at, um, uh, up at University of Pennsylvania, up at Brigham and Women’s Hospital up in up in Boston. Um, you know, these, these are serving kind of multiple missions at the same time. They’re trying to deliver care, but they’re also advancing research.

Rick Abramson: Um, a lot of academic medical centers are kind of interested in developing their own technologies. Um, Um, either, you know, for research purposes or potentially for commercial purposes as well. Um, and, um, and I think that they’ve had mixed success in that area. Um, you know, some, there’s some really great success stories of new technology split spinning out from, from academia, as you know.

Rick Abramson: Um, there’s also, um, [00:15:00] a lot of, uh, you know, unfortunately, kind of a history of, um, uh, creating, um, solutions to problems that aren’t really much Of problems, you know, if that makes, if that makes sense, you know, um, elegant solutions to, uh, to doing something that nobody really is bothered by, um, in the first place.

Rick Abramson: You know, I, I think that’s a challenge that’s faced by, by academic centers, but at the same time, they, they, they’re serving an important role of, of, of furthering our basic knowledge, our basic understanding. And so I think that that’s, um, you know, that’s, that, that’s something that every academic medical center at Russell’s with, um, you know, HCA, um, was an experience and I’m, I’m a couple of years out at this point.

Rick Abramson: point. Um, but, um, I can say that what was going on at HCA when I was there, uh, was HCA trying to figure out how to capitalize on this vast, vast store of extremely, extremely valuable data, um, that it was sitting on just as a, as a large organization. And I think that’s a, that’s, you know, it’s from an investment standpoint, you know, I think that’s actually an important, important, uh, you know, kind of, uh, Uh, thing to underline, which is that, you know, [00:16:00] scale has advantages other than just, um, uh, economic resource.

Rick Abramson: Um, scale has data resources as well. And so there are certain things that a organization like HCA can do in terms of developing new applications and new solutions, just by virtue of its enormous size and enormous access to data, um, that a smaller institution could not do, even if it had the, the just bare economic resources to do it.

Rick Abramson: Um, now when I was at HCA. A lot of the focus was not on, uh, clinical AI, but rather on, um, operational AI, if you will, um, so AI for, uh, to kind of streamline, uh, operations in the hospitals, in the facilities, um, resource management, um, how to move patients through workflows, um, uh, how to optimize care pathways, but not from a clinical standpoint, but rather a non clinical standpoint.

Rick Abramson: Um, and, um, and that was really interesting for me to see really, uh, something, you know, from, as a, as an impartial observer to, to, to watch. Um, and it made a lot of sense, um, when I, when I saw it happening, [00:17:00] because it’s an organization that, um, is, uh, you know, responsible to shareholders. It wants to, um, capture efficiencies.

Rick Abramson: It wants to kind of go after the low hanging fruit. Um, and, um, and this is, um, you know, something that’s reflected in the larger market. So let me now kind of shift to the larger market and the, the startups and the kind of the VC backed entities that you, that you spoke of, um, when I look at AI, um, and this is.

Rick Abramson: beyond medical imaging. This is like all of kind of AI for health care. Um, the first kind of branch point that I have when I kind of look at this universe is between clinical applications and non clinical applications. Um, and so, you know, in, for, for us, for in radiology, the, the clinical applications are the, those, that’s the sexy stuff, right?

Rick Abramson: That’s like, oh, we’re going to create a software that’s going to read your x ray, you know, that’s like, everybody gets really excited about it. But, um, but any discussion of AI for medical imaging is woefully incomplete. If you don’t, um, if you don’t talk about AI for non clinical aspects [00:18:00] of imaging, and this is the boring stuff and it’s the non sexy stuff, but it’s actually really, really important.

Rick Abramson: It’s the prior authorization and it’s the scheduling and it’s a staffing and it’s the charge capture and the revenue cycle management and the business analytics, right? Um, and obviously you guys know this as investors, but a lot of people who who don’t think about this ecosystem, they automatically jump to the sexy stuff and they forget like this huge value chain, you know that exists before and after the delivery of the actual clinical service.

Rick Abramson: And that’s actually where I arguably I think a lot of the attention is focused right now. Um, so, you know in that VC startup world, Um, a lot of the more successful entities that I’ve seen, and it would be curious if this resonates with with your, um, with your experience as well. A lot of the more successful plays have been non clinical in nature.

Rick Abramson: You know, going after the, you know, boring, but really important stuff and, uh, and optimizing processes that are, um, non clinical, um, exempt from, regulatory oversight, which [00:19:00] is really important. We’ll probably come back to that. Um, and, and more importantly, stuff that’s just really, it’s, it’s just, it’s dollars and cents.

Rick Abramson: It’s, it’s efficiency, it’s operations, it’s like the stuff, this block and tackle stuff, um, that you have to kind of do. And it’s a lot, it’s like easier to accomplish than a lot of the more ambitious clinical stuff that we have, that we have going on. So, um, so like that’s, that’s kind of how I’m looking at the, at the ecosystem and, um, uh, and, and how I’m, I’m kind of looking at, uh, kind of like how these different entities have, uh, have kind of approached the problem.

Vic: Yeah, yeah, I think that’s, that we have the exact same philosophy. I think there’s sort of two different, uh, levels of adoption maybe, and while it may be boring, It’s about 20 percent of something like a 5 trillion annual spend. And so you’re looking at about trillion dollars in overhead. So even the boring stuff can be a significant number of, uh, pennies there.

Vic: But I think the, uh, the natural place that is, [00:20:00] uh, really impactful to care And also can be, uh, somewhat daunting and emotionally triggering to some folks, patients and doctors, is moving into the, into the clinic, into actually the delivery of care, because that’s the other 80%. So certainly the first movers and where a lot of the adoption has happened, where a lot of the venture money has gone.

Vic: is in that first non clinical revenue cycle, maybe helping with, uh, physician notes, all, all of that stuff that is adjacent to clinical. Um, and I think there’s a lot of patients, a lot of doctors that I think are excited about the AI opportunity. In treating, in changing the way we care for patients, and there’s other people that are nervous about that for privacy reasons and all kinds of reasons that we all understand.

Vic: So I think it is important to differentiate that, um, maybe give me a sense, give us a sense for what [00:21:00] your view is in terms of like, forget the regulatory apparatus, but just, just the pure technology capabilities today. Uh, as it relates to radio or wherever you want to take it, like, where is the state of the technology today if we didn’t have to deal with regulation?

Vic: Of course, we’ll get to that in a minute, just how, how functional is the technology? As you know as a radiologist looking into this stuff. Yeah, where is it useful? Where is it not yet?

Rick Abramson: Yeah. Yeah Um, okay, so i’m bullish. Uh, let’s start there. Uh, you know, i’m an optimist, uh, and you know kind of coming out of this This kind of immersive experience that i’ve had for the last three or four years Um, I think the technology is really good and it’s getting better Uh, at a pretty good clip.

Rick Abramson: Um, and, um, and let me give some specifics because it’s it’s easy to just kind of hand wave and say, like, Oh, we’re doing great things. But, um, let me tell you kind of what I’m seeing across the globe. All right. So I’m seeing, um, in medical imaging, I’m seeing software being deployed, um, with a specific intention of [00:22:00] catching errors and catching misses.

Rick Abramson: Okay. And so these would be software for, uh, that would kind of scan, uh, scan an image as it’s being interpreted by a radiologist and say, uh, like would say like, Hey, wait a second, you know, radiologist, there might be a little spot on the lung that you may have missed. Yeah, you may have neglected this little spot on the lung, and that actually might be an early lung cancer.

Rick Abramson: So that’s like a very, very specific concrete example of an application that we’re seeing, and that’s, that’s real, real world today, you know, in 2024, we can actually do that. Um, we can say, hey, look, there may be a, um, an early stroke. Um, that you’re that you’re not that you that you neglected that you missed that were that you haven’t picked up.

Rick Abramson: Um, a little hemorrhage in the brain. Um, it’s going to dramatically change care for this emergency room patient. Um, we’re seeing that again. That’s that’s real world in 2024. We’re seeing that in use right now. Um, we’re seeing technology that will flag urgent cases. [00:23:00] Um, in the E. R. and pop those urgent cases up to the top of the work list so that they get interpreted sooner communicated to the referring team sooner and that care is initiated more quickly.

Rick Abramson: Um, that’s real world right now. Um, on the other end of the. Spectrum, you know, as far as kind of like of care delivery. Um, we’re seeing a I being used to screen populations for infectious disease. Uh, we’re seeing, um, uh, a I tools being deployed to screen for tuberculosis in developed countries. developing countries.

Rick Abramson: Um, we’re also seeing AI being used to, uh, screen and help screen for breast cancer, for example, and, and for screening mammography. Um, we’re seeing cases of AI being used to intelligently triage care to specialists. You know, for example, this is something that, that should go to neurosurgery and we’re going to expedite that triage to neurosurgery.

Rick Abramson: Um, and we are very, we’re just beginning. just beginning to see AI being used to draft the [00:24:00] reports for radiologists. Um, and I put that in kind of a special category and it might be something to come back to later because, um, this starts getting into the use of some of the more advanced general AI like, uh, chat GPT type technology, like these large language models.

Rick Abramson: Um, and there, you know, there’s been a lot experimentation about kind of like whether these Language models can be applied to radiology. Um, I think the jury is still out. Um, and for a bunch of reasons, I think it’s a little bit premature, um, to start thinking that chat GPT is going to replace radiologists.

Rick Abramson: Um, but, um, but it’s an, it’s an example of, of, of what people are thinking about, um, in, in terms of, uh, in terms of the technology. Um, Now, one thing I will say, um, Vic, just to kind of answer your question and kind of do it fairly is to say that, you know, I’m, I’m giving examples that are kind of empiric in nature, right?

Rick Abramson: And I’m just saying I can, I can point to anecdotes of a missed lung cancer that AI caught or a missed, you know, fracture that AI caught. Um, but it’s one thing to point to [00:25:00] anecdotes, it’s another to back up your statement with statistics, right? And I’d say that the, the, the clinical evidence, um, for this stuff is, is, is, is here.

Rick Abramson: It’s present, but it’s still somewhat rudimentary, um, and that’s for a variety of reasons. Um, uh, we can measure the effect of AI on physician performance, but the data gets a little bit, um, it’s noisy and, uh, it’s, it’s a little bit nebulous. We need better trials, larger trials, larger studies of, of the, of the technology to make sure that, that, um, that we’re not just relying on, on anecdotes.

Rick Abramson: And, uh, and just one off examples here and there. Um, but I would say in my, in my experience, I mean, I can certainly point to those anecdotal examples. Um, and I can say that the models are only getting better and better.

Vic: Okay. Excellent. And just to help me understand it, maybe help the listeners. Are there a couple of different categories that you framed up?

Vic: We use an example. So. Maybe a population being screened for tuberculosis in a [00:26:00] developing world that we’re not doing today, that now with these tools, they’re lower cost, you could get screenings to areas that are not possible for. That seems like a pretty easy use case. Then the next one is, uh, modifying the triage, either how the images are read, or pushing referrals to the right specialty.

Vic: They were sort of an ED use case and then also a subspecialty referral. That’s, that’s more of a subspecialty referral. I don’t know, triage appropriately, or, or use AI to organize the caseload more effectively. Again, that seems pretty straightforward. And then you start to get maybe a little more, uh, anecdotal need to do more research on, which is, finding a concerning spot in the image that maybe the human doctor didn’t see, um, that could be really valuable, but, but [00:27:00] maybe it is a false positive that is not.

Vic: So that’s maybe where there’s a lot of promise, but it’s hard to know the net effect over a large population without a study. So that’s maybe another category. And then the last would be, Automating some of the physician’s work, doing a report, which is the furthest, um, unknown, but has a lot of potential.

Vic: Is that, is that a fair categorization of the different examples?

Rick Abramson: Yeah, no, I think that’s a good way of breaking it out. Um, you know, you could even do it more simplistically and just talk about quality and efficiency as being kind of the two arms, you know, the two levers that we’re trying to, that we’re trying to move.

Rick Abramson: Um, with with efficiency being both efficiency of care delivery and efficiency of physician work. Um, but if you’re kind of looking for, I’m kind of, um, I’m anticipating kind of like the next, the next step in the conversation, which is going to be about potentially about ROI. Um, you know, those would be kind of like the, the very, very high level ROI levers that I think that we’re, we’re trying to get at, if you look at it kind of from a, uh, from a [00:28:00] business model, business model standpoint.

Vic: Right. And that’s where some of the public health developing world, um, if someone’s willing to pay for it, that’s a, that’s very good to bring more screenings to develop world. Uh, the ROI is maybe, uh, not as carefully looked at because it’s a, it’s, it’s some nonprofit maybe that’s trying to get better public health out there, which is probably different than how a health system would think about organizing there.

Vic: Right. Their human talent to get the most efficiency out of the out of the patient population.

Rick Abramson: Yeah, the whole conversation is tied into the question of, um, you know, just kind of, you know, business model development and who is the ultimate customer. Uh, and, and, and that, that in itself is, is, is something that the AI industry is, I think, is still wrestling with, uh, kind of defining on a very, very fundamental level, you know, who, who is the ultimate customer and all this.

Rick Abramson: I mean, the ultimate customer is the patient, right? But I mean, in terms of who’s, who’s writing the check. Uh, you know, [00:29:00] who who’s kind of funding this development? Is it the provider? Is it the hospital? Uh, is it government? Is it non government organizations? Uh, is it pharma? You know, there’s kind of like you can kind of think about it.

Rick Abramson: There’s actually uh, there’s stuff going on right now where pharma is being leveraged to in service of Global health, right? Because like pharma, uh, you know, pharma wants to get out drugs for early lung cancer. And, um, uh, and so they potentially funding AI in for use in a large scale, uh, global public health initiative.

Rick Abramson: Um, so it’s, but instead of funding it through an NGO or through the world health organization, it’s actually funded by, uh, by private pharma. Um, so that’s a model, uh, payers is another potential customer. Um, you know, and so, you know, there’s kind of different value narratives associated with different customers.

Rick Abramson: And, um, and we’re so early in this industry that I can tell you all of these models are still on the table and they’re still being explored. Uh, and I don’t think we’ve, uh, we’ve nailed down yet what the ultimate model is going to be. [00:30:00]

Marcus: Just, just more of a, uh, a question around. What it’s like to be, you know, an executive with a fair amount of experience in, in this, in this industry who was watching AI before LLMs came around, you know, I, I heard you say LLMs, uh, replacing, uh, radiologists, probably a little bit premature.

Marcus: And I think we, we’ve all got the question around when hallucin hallucinations are going to You know, get to a place where we’re all a lot more comfortable about handing over the keys to these machines, which, by the way, none of us know how they actually work. Not even the people who created the transformers can actually explain how they work on the inside.

Marcus: You know, it sounds like when you talked about your different, um, Anecdotes of of where AI is making an impact today in patients lives that those were not LLM examples. Um, you know, they were more sort of machine learning examples, which was kind of what we used to talk about when we talked about AI before Sam Altman came on the scene.

Marcus: Um, [00:31:00] so I’m just sort of wondering now that. AI has largely become synonymous with LLMs, uh, since chat GPT has, has broken out on the scene. Uh, how, how confusing is it for the general public as you, uh, are seeking to build relationships, build partnerships, you know, talk with people who are not as, uh, deep in the, in the space as you are around what.

Marcus: AI means in its broadest sense and really in the, in the areas where you believe it can truly create value today versus LLMs, which hold unbelievable promise for the future, but today have a lot of barriers to include a regulatory threshold when we’re talking about them, replacing clinicians.

Rick Abramson: Yeah, it’s such a great question.

Rick Abramson: And it’s all wrapped up in this very, very intertwined, intertangled story about regulatory, about ROI, um, about, uh, and about just how the general market understands the technology. So, um, I mentioned before that I [00:32:00] kind of, when I look at the AI universe, I’m always thinking, kind of thinking in terms of clinical versus non clinical, uh, the other kind of big branch, uh, you know, kind of like kind of categorization that I always do is between general AI and more narrow AI models.

Rick Abramson: And that’s exactly, I think what you were referring to, right. In terms of kind of chat GPT being the exemplar of a, of a general AI tool. And then kind of what I’ve been talking about is more, you know, very narrowly focused, you know, specifically trained, uh, more machine learning applications being more of the, the narrow AI and, uh, and the conversation completely changed in November, what, November 22.

Rick Abramson: Is that when I think, uh, when chat GPT was released, um, because, uh, up until then, it was all just the narrow machine learning models. And now all of a sudden we’ve got this general kind of general intelligence. a general, you know, large language model, uh, you know, entity to deal with and to ask, you know, how can, how does this fit into the whole, whole equation?

Rick Abramson: Um, you know, for us in radiology, the ultimate application of the LLMs would be, [00:33:00] as I said before, to draft the report. Now, why is that so important to radiology? Well, it’s because Because the unit currency in radiology is really the report. I mean, that’s the product. You know, if you kind of think of it in terms of it in an industrial mindset, I mean, the pro the end product that I create as a radiologist is a, is a report, uh, is either a slip of paper or electronic report, but it’s, it’s electronic report and it’s text.

Rick Abramson: So, you know, if you can get an LLM to generate that text, well, all of a sudden that’s a huge shift in my workflow and potentially an unbelievable shift in my efficiency as a, as a producer. And the reason I say it’s kind of wrapped up in the conversation around ROI is because, you know, like keep in mind that like this technology by and large across the globe is not yet reimbursed.

Rick Abramson: This is not a reimbursed technology. There’s no CPT codes. I mean, there’s a couple of CPT and CPT codes here and there, but for the most part, This is not reimbursed. And so if I put on my sales hat and if I go to a hospital or radiology group, um, [00:34:00] it’s not a matter of just convincing them to adopt my technology.

Rick Abramson: I have to get them to pay for it. I got to actually sell this technology to them. And, um, you know, unfortunately, or fortunately, depending on your perspective, you know, quality improvement or quality maintenance in and of itself. It’s hard to sell. You know, like it’s hard to sell quality, especially in the United States, um, without an associated efficiency improvement.

Rick Abramson: And so that efficiency gain becomes extremely, extremely important. Um, now we have some data. I had some data from, uh, from my time at Annalise, um, you know, suggesting that even these narrow machine lean machine learning tools, um, Uh, may actually make radiologists more efficient, uh, slightly more efficient.

Rick Abramson: And when I say slightly, I mean like, you know, maybe 10 percent more efficient, you know, 15 percent more efficient, uh, which is nothing to sneeze at. Uh, but then, uh, y’all have lunch with a friend of mine in private equity. And they say to me, you know, so, you know, we want to do a big roll up and we’re going to open up a can of AI.

Rick Abramson: And, you know, just tell me, am I looking at like a three X efficiency [00:35:00] gain, four X gain, five X gain? I’m like, Hmm, how about like a 1. 1 X gain? And then, and then lunch is over pretty, pretty quickly. And then I have to, um, so, you know, it’s an incremental improvement with these narrow tools, but with a general tool, a chat GPT, man, if these things are actually.

Rick Abramson: drafting the reports for me, and all I have to do is just glance at the image, glance at the report, make sure it’s right, and sign off. I mean, wow, that’s, that’s huge. I mean, now we are talking about 3x, 4x, 5x gains, and that’s really, really gets people attention. The problem is, and you probably know this, the problem is the, these chat GPT, GPT models, they’re not just making up and fabricating references in college, you know, book reports, they’re also hallucinating clinical Uh, you know, clinical findings as well.

Rick Abramson: Um, and I don’t think anyone has yet been able to prove that these models can generate hallucination free reports or fabrication free reports at the level they would need [00:36:00] to be in order to be incorporated into clinical workloads. Um, and so we’re, we’re, we’re stuck for now. Um, and I think that the big debate in going and going on in the academic circles and also in industry is whether.

Rick Abramson: You know, I mean, everybody wants to automate reports. It’s like, how do we do it? Are we going to refine chat GPT to do it? Or are we going to take one of these kind of more narrow machine learning approaches and kind of put some kind of text generation overlay on top of that? That’s where the, I mean, that’s really kind of bleeding edge, I think, in terms of the technology development.

Rick Abramson: But that’s the holy grail in in, in radiology is automating that report. Um, and then again, and so then, okay, now let’s, let’s open that back up, take it outside of radiology and look at all these other clinical disciplines. So we asked ourselves, we would ask ourself, okay, what is the ultimate product? You know, um, Um, you know, in, in, uh, urologic surgery, maybe the pro the ultimate product is a prostatectomy.

Rick Abramson: Okay. And so, oh, we’ve got robots to do our [00:37:00] prostatectomy. Great. How much of an efficiency improvement is that for urologic surgeons? So you kind of, you start using the same, I mean, not the chat GPT is going to start doing. Robotic prostatectomies. But I mean, it’s the same. I think, like, when we look at AI in terms of, and ask ourselves whether AI is going to present enough of an ROI to promote large scale adoption in absence of reimbursement, that’s the kind of mentality I think we have to, we have to use.

Rick Abramson: And it’s a pretty high hurdle. It’s a, it’s a very, very high bar. Um, and so, um, and so, and I think that’s why a lot of clinical AI startups continue to struggle. Because they’re showing quality improvement. They’re showing incremental efficiency improvement. Just not enough to, you know, to, to, to make the bean counter say like, okay, yeah, this is something we, we have to do.

Rick Abramson: Yeah, this is a, uh, a must do rather than a nice to do. Right. Yeah.

Vic: Yeah. So that’s where I was going to go next. Healthcare is You know, at least in the U. S., well, actually, globally, highly regulated, it’s regulated different ways based on the country, um, [00:38:00] and I think the regulators probably rightly have a duty to ensure safety and reliability before they start sort of opening up the opportunity.

Vic: What’s your view of, uh, how, uh, AI, whether it’s narrow AI or broader LLM, general, general AI is regulated today. And then where do you see the regulation? Kind of going to the US and then globally.

Rick Abramson: Yeah. Okay. Well, so, um, i’ll start one more time with that clinical non clinical distinction Um, and that’s important for regulatory because if you can escape Um fda regulation as a clinical device Uh, your entire world is different your entire go to market your entire sales and marketing strategy is completely different Than if you’re regulated as a clinical device and I can point to you know, several You several startups in my space, um, whose entire business strategy revolves around making sure that they’re not classified as a [00:39:00] clinical medical device by, by the FDA.

Rick Abramson: Um, and so that’s, that’s very, that’s very important. I think just first to understand that like, if you, Know if you, if you can be nonclinical, you are in a much different place than if you’re a clinical. Now if, if you are clinical, uh, you’re regulated as a medical device, um, in the US would be by the FDA and the, in Europe it would be EMA with the CE mark Australia, it’s the TGA, the Therapeutic Goods Administration, and Japan as A-P-M-D-A.

Rick Abramson: But it’s all, um, under the set of regulations that apply to medical devices. And so the same, uh, general, uh, locate, general outfit within the FDA, uh, that regulates, you know, cardiac defibrillators and, uh, catheters and tongue depressors, uh, it’s the same, that, that same part of FDA, uh, that regulates, uh, AI.

Rick Abramson: Um, now, um, FDA has been, been really scrambling of the past several years to, to modernize its, uh, regulatory framework. Um, You know, my [00:40:00] opinion right now, and I’ll I’ll preface this by saying that I’m not a I’m not a defund the FDA guy. Like, I don’t want to blow up the FDA. I think the FDA serves an extremely, extremely important purpose.

Rick Abramson: Um, you know, making sure that, uh, the drugs and devices out on the market are safe and effective and really, really safeguards the American consumer with that. As you know, as a, as a, as a preface, I think that the FDA, um, I think is probably at this point overly stringent with its regulation of AI devices.

Rick Abramson: And I think is, uh, impeding innovation, um, and, and access to innovative, uh, uh, innovative products in the American marketplace. And I say that because I say that not with without evidence to back it up. And we can point empirically to, um, uh, approved AI devices in Europe and compare it to approved AI devices in America.

Rick Abramson: And, uh, Europe is, is, is, um, uh, leads the pack and is moving further and further away. In other words, there’s a growing, uh, I would call it innovation gap between Europe and the [00:41:00] US. Um, and which means that European patients have access to, uh, you know, to more. More cutting edge products than than American patients do, um, now that’s obviously it could be interpreted either way.

Rick Abramson: I mean, it could be you could interpret it that that Europe has just gone off the rails and they’re just approving, uh, you know, stuff willy nilly without without proper, you know, evidence of, um, of efficacy and safety. Um, I think the big difference is what Europe has done. It has a Europe has adopted more of a mindset of, uh, get products out to market.

Rick Abramson: And then perform post market surveillance to make sure that those products are safe and effective. And if not, remove them from the market, whereas America, the FDA is doing the emphasis is more on pre market evidence and the evidentiary burden on developers trying to come into the U. S. Is, is very high, you know, so high.

Rick Abramson: In fact, that you see a lot of companies looking at global companies, um, looking at the U. S. market saying [00:42:00] great market. Would be so nice if I got, if I could get in, but it’s just, it’s just too expensive. And I’m just going to stay out of the U. S. market. I’m going to sell my products elsewhere. Um, either that, or I’m going to kind of sell some dumbed down version of my product into the U.

Rick Abramson: S. You know, something that I know that I can get through regulatory, uh, but not, not a product with the full functionality, um, that I deploy, uh, elsewhere. Um, which is, you know, again, like from my perspective, that that’s that’s unfortunate. Um, you know, others could look at that and say that, you know, that U.

Rick Abramson: S. is being appropriately prudent and, um, in terms of kind of having this higher level, um, of, uh, uh, these higher evidentiary standards. Um, but, you know, look, the way I’ve come to look at this whole area, I kind of look at this, you know, especially for medical imaging, but elsewhere as well, I’m looking, I look at AI primarily as a, as a safety device.

Rick Abramson: Yeah, that’s, that’s kind of like the, the, the, the paradigm that I’ve adopted, you know, going back to my empirical examples of [00:43:00] AI, you know, catching a missed lung cancer or catching a missed fracture or missed stroke, um, you know, I’m looking at AI primarily as a second pair of eyes for me. Uh, because I’m stressed, I’m overworked, I’m, I have too many studies to read.

Rick Abramson: Um, AI is kind of like something that’s operating in the background to make sure that I’m delivering high quality care. Um, and in my mind, you know, maybe it’s because I have a teenager who just got his driver’s license that I’ve, that I’m starting to think this way, but, um, you know, Alex, my 16 year old, he’s been driving now for a couple of months.

Rick Abramson: And he’s out on the road now with a bunch of really stressed out drivers who are, you know, like the roads are busier now than ever before. Uh, you know, it’s very similar to kind of like the procedural environment in healthcare. Um, I sleep a little bit better at night knowing that he’s driving a car that’s got like the little, you know, the, Lane shift monitor and the blind spot protection and the little thing that beeps if he’s going to back up into a fire hydrant, that type of thing.

Rick Abramson: And, you know, I, I kind of, I kind of think of, of AI for radiology in the same way that I would actually [00:44:00] sleep better as a radiologist and as a patient as well. And knowing that they have these safety devices working in the background. Um, and so for me, that means kind of a lower, you know, slightly lower burden of, of clinical evidence, you know, not dropping the burden entirely, but, um, you know, just making sure that that regulatory burden is commensurate with the level of risk.

Rick Abramson: Um, and right now I think it’s, it’s disproportionate and I think it’s, it’s ready for a readjustment.

Vic: And where is the line in your mind? Is it, um, That the doctor needs to make the diagnosis and care plan and the AI should be a tool providing information or highlighting a part of the image and that would be okay or is aligned somewhere else like what is what needs to be approved differently versus what can be brought in.

Vic: Under the existing regulations.

Rick Abramson: Yeah, it’s okay. So that’s that’s really interesting because that gets it to the this kind of, uh, [00:45:00] like a yet yet another taxonomy, right? Yet another way of categorizing these things. But, um, if you the A. M. A. Has actually come out with a framework. It’s actually supposed to be used not so much for regulatory, but actually more for reimbursement.

Rick Abramson: But, um, they categorize these A. I. Tools as assistive, augmentative or autonomous. Okay, so assistive. Okay. As it implies, it’s assisting the physician, but the physician is still in the driver’s seat. You know, it’s kind of the physician’s doing their job. It’s just you get a little bit of AI help, you know, on the side.

Rick Abramson: Augmentative, the AI is actually, you know, the physician is still in the driver’s seat, but the AI is actually providing some information that the physician may not be able to generate. His or herself, him and him or herself. So it might be might provide like a risk score. For example, that would be an example of something that if I’m just looking at a lung nodule, I can’t say just with my own clinical knowledge, the percentage chance of it being cancer.

Rick Abramson: But if I have an algorithm that actually has a nomogram and has [00:46:00] refers to some clinical trial data and actually say, Oh, it’s a 63 percent chance of it being lung cancer. That would be that second category augmentative because it’s providing some additional information that I wouldn’t otherwise be able to generate.

Rick Abramson: And the third category is probably the most interesting and the most vexing, you know, for, for kind of like, you know, just kind of thinking about things is that autonomous, autonomous AI, uh, where the, where the AI is actually replacing the physician and actually doing stuff like, as it says, as it implies autonomously.

Rick Abramson: Now, um, the fascinating thing to me, I mean, just when you talk about kind of learning in these different roles, I mean, as the chief medical officer for an AI company, the most, one of the most fascinating learnings for me is that. Autonomous AI actually probably has an easier pathway than assistive AI, which was completely counterintuitive to me when I kind of first started wrestling with it, but now it’s starting to make a little bit more sense.

Rick Abramson: This idea of physicians being assisted by AI gets you into a whole bunch of just [00:47:00] metaphysical issues and philosophical issues about what the technology is doing to the physical world. to the physician? Is the, is the technology augmenting the physician’s judgment? Is it potentially biasing the physician’s judgment?

Rick Abramson: You know, what are the, you know, the fact, we know that none of these AI tools are perfect. And so these false positives, these false negatives, uh, it’s a real conundrum of kind of thinking about kind of, okay, how is the physician as the primary operator transformed from, Not using the technology to working with the technology, and I don’t think anybody has really kind of gotten there, like kind of really wrapped their brain around like all these implications, which is why I think all the regulators struggle with it.

Rick Abramson: If you look at autonomous AI, though, that’s actually a much simpler problem. Because you don’t have to look at the effect of the AI on a human operator. You just have to look at the AI’s performance. And we have statistics to tell you that. I can tell you, like, an autonomous, um, mammography engine. I can, yeah, I can run a very, very simple clinical [00:48:00] trial and tell you the accuracy of that software tool for identifying a breast mass.

Rick Abramson: But it’s a lot harder to actually say, okay, what are, if you have a mammographer using the tool, you know, what are the effects of the, on the, the tool on the mammographer? And so there’s, it’s actually paradoxically, I think, um, I think going to be easier to have autonomous tools. Uh, get through regulatory and to get reimbursed, uh, than the assistive tools.

Rick Abramson: Now we’re not ready for the autonomous tools yet, but when they get here, I think they’re actually, they actually may actually, uh, they may actually lap of the assistive tools, um, in terms of making their way to market.

Marcus: That’s that’s

Vic: interesting. It is counterintuitive.

Marcus: Yeah. And also like not. Like the conversation we’ve been having in previous.

Marcus: So we we’ve had conversations with, uh, Tarun Kapoor, who’s that, uh, virtual health, uh, my friend, uh, Ambar Bhattacharya. He’s a VC at Maverick, um, talk to both of them. And they were sort of talking about the paradigm of co pilot versus autopilot, right. You know, and, [00:49:00] and which is what you just sort of talked about.

Marcus: And generally speaking, we’ve been talking about how, The public is can, can better wrap their head around co pilot than they can around autopilot, just because even if the outcome of the autopilot is better. On the whole, the minute the autopilot fails, it’s, you know, the machine killed a person, right? And, and so you get into the whole kind of like Frankenstein chase them, you know, over the hill kind of thing.

Marcus: Um, but I, but I think in a regulatory context, what you’re saying does make, does make some sense. And I think we, we, I think the other thing that’s, that’s sort of, um, Uh, corroborating your point is we are now having a very broad conversation about the ways in which technology is impacting humans from a mental health perspective, uh, by way of our conversation about social media right now, you know, like we let social media run [00:50:00] unfettered for basically a decade, uh, you know, before we actually stopped and said, Hey, Has anyone looked at this thing and studied it to see whether or not it’s actually impairing the way we are thinking and feeling?

Marcus: Um, and now that that conversation is front and center, uh, you know, I, I think you’re right. I think we’re, we’re now going to be looking at the way that technology does influence people in all manners of, of life. And, uh, that’s, that’s interesting.

Vic: Yeah, and I think it is, um, it just demonstrates the challenges that the regulatory people, the people in these agencies have to wrestle with, right?

Vic: Like, it probably is an easier statistical challenge, right? To test the fully autonomous AI tool compared to the status quo. And at some level, it would be sufficiently good quality That it would be it could be possible to allow it and yet with your [00:51:00] son and my driving and my sons And we’re all used to the co pilot um cruise control or the lane change or you know, we fly on airlines and they all have autopilot Uh, but there’s a there’s a pilot sitting there to make sure they land and take off and everything if there’s any problems So the public perception is just like I think the human Understanding of it.

Vic: It seems more scary to be fully autonomous, but it probably is easier to control as a regulator. So you have this public perception versus the, the statistical analysis.

Rick Abramson: Yeah. And I did, and I do want to just reemphasize that I was really talking about regulatory and reimbursement very, very narrow. Yeah. I completely agree that the larger market adoption, I mean, that, that significant headwinds, right.

Rick Abramson: In terms of kind of public’s willingness to, uh, to, to, uh, uh, to endorse and adopt, you know, the autopilot. Model for healthcare. Although, you know, I mean, [00:52:00] maybe, maybe we should work on that, right? Because if you use the, uh, self driving car as the, you know, it’s always the analogy we go to. I mean, that, you know, the imagine a world where it’s all autonomous cars.

Rick Abramson: Um, you know, I think that 45, 000 deaths a year from car accidents. I think that number would be orders of magnitude lower. Um, but it’s not going to be zero. And I think, yeah, I think to your point, the, the, the public is not going to tolerate those, you know, non zero deaths from autonomous cars on for, even, even if it’s like a, a huge public health, uh, benefit by, by making the shift.

Vic: Right, right. Look to the future a little bit. I mean, you’re. in a startup, uh, talk about that technology that you’re, uh, working with, but then also help us understand what should we be watching for in the next three, six, 12, 18 months? What, what are the big technical technical advances that you think could really unlock significant [00:53:00] market opportunities?

Vic: Um, and you may be in one of them, so maybe an Anis ai, uh, probably not saying that. Right. How do you, how do you say the company, oh, uh, anise, Annelise, Annelise AI may be, may be a poster child for that. You may be an example of that. But just give us the overall view of technology.

Rick Abramson: Sure, sure. Yeah. Um, and I should say that I’ve actually, I’ve actually.

Rick Abramson: Pulled away from my full time role at Annalise Effective just a couple weeks ago, just for other factors, complex factors that we don’t have to go into, um, mostly logistics around working for a company in Australia. In Australia, yeah. Yeah, yeah, yeah. I continue to serve them as a consultant, so, um, so I’m not fully away from them.

Rick Abramson: But, um, but yeah, so, okay, let me, um, let me talk about, uh, I’ll talk about radiology, but let me talk about the larger, uh, Um, AI landscape as well. And again, I’d love to see if this, this, um, uh, you know, how this dovetails with, uh, with, with your experience as well, kind of talking to, to entrepreneurs. Um, so I think the, the big, uh, next 12 to 18 months for, uh, for radiology medical imaging, um, there’s going to be a lot [00:54:00] more discussion around LLMs.

Rick Abramson: Uh, so I think let’s, let’s start there. Um, uh, it’s not all gonna be about, uh, drafting reports and, you know, how best to do it, whether to use, uh, you know, these general, uh, large language models or more, or kind of, you know, versions of, of, of narrow tailored, uh, machine learning. Um. There is, um, uh, I think everyone is very much waiting to see if the FDA, uh, changes its approach to, to regulation, um, potentially after the November election.

Rick Abramson: I think there’s, you know, kind of a lot, there’s going to be a lot of, uh, let’s, let’s, let’s watch the FDA and see kind of what, what comes out. Um, there’s also, um, a phenomenon that’s kind of emerged over the past maybe 12 to 24 months, um, uh, called foundation models. And, uh, these, uh, these exist not only obviously in imaging, but elsewhere as well.

Rick Abramson: Um, these foundation models, essentially, I think the best way I can describe them is to say that these are these large kind of pre trained algorithms that are like 85 percent baked. Uh, they’re kind of like [00:55:00] 85 percent ready, uh, because they’re used these large data sets to make these really, really great models.

Rick Abramson: Um, but there’s a, but there’s a little bit left undone so that you can come into a site, you can come into an HCA or a tenant, or you can come into a Cleveland clinic or whatever. Um, and then you can train the rest of the model on local data, on local data from the, from the system, um, with the idea that that model then kind of becomes, More narrowly tailored to that, uh, to that site to that system.

Rick Abramson: Um, uh, potentially less, uh, biased because it kind of incorporates more local data. Um, and it’s also something that the, that the, that the site has some partial ownership over. Um, So I think those are going to be very important. Um, you know, not just in, um, medical imaging, but in other areas of, of kind of AI and data science and healthcare.

Rick Abramson: Um, and so I’m going to be watching that very carefully, both from a business perspective and also from a regulatory perspective as well, because it has some kind of, there, there are some, some interesting intersections [00:56:00] with, uh, with the FDA, um, apparatus there as well. Um, and then let me, I’ll just try to kind of bring it outside of imaging, um, and, um, you know, everyone’s very excited about AI and healthcare, especially for clinical use, but, um, uh, oftentimes there’s a bunch, it’s kind of vague, you know, people say like, Oh, we’re going to have these computational insights and then it’s kind of like, well, what, what, what are those computational insights?

Rick Abramson: And so, um, Let me offer some areas that I think, um, are interesting to me, um, and, uh, to see like, kind of like how, how, how these, um, mesh with what you’re seeing. So, um, diagnostic medicine, uh, is the first area. So not only, um, medical imaging, but also pathology and dermatology, anything, any part of medicine that’s, uh, kind of, um, uh, relies on, on visual pattern recognition, I think is, um, very much amenable to these machine learning techniques.

Rick Abramson: And I think, Uh can and will be exploited. Um, the pathology, I think the rate limiting step is the [00:57:00] adoption of digitization. Uh, but once, once we kind of like have a big shift over into digital pathology, um, that’s potentially going to be even, I think even bigger than radiology, uh, to be honest. Um, so I’ll be, I’ll be watching that very carefully.

Rick Abramson: Um, I think so, um, uh, for procedural medicine, um, I think, uh, I’m, I think we’re, we’re still at the tip of the iceberg in terms of robotic. Um, robotic approaches, uh, to, to, uh, to procedures. Um, so I’ll be watching that. Um, I think, um, Vic, you mentioned, I think earlier you mentioned, um, I think you alluded to NLP for clinic notes.

Rick Abramson: Um, I think that’s, that’s, that’s something that’s, that’s, so now. NLP being natural language processing your listeners probably are familiar with but um, uh, that’s really kind of taking the world by storm right now Um, and um, and again, it’s one of those areas that it’s it’s low hanging fruit and it’s real dollars and cents if you can speed up the patient kind of throughput in a clinic Um, that’s really really important for for a health system.

Rick Abramson: Um, and so I think we’re going to see, you know, [00:58:00] greater adoption there um, and then I think the Final area that I’m really excited about in health care in AI is drug discovery, just, uh, you know, kind of targeted drug, drug discovery and, and, and for, uh, for, for therapeutics, um, which is a little bit outside of my wheelhouse, but, um, uh, but I still think all the same considerations that we, that we, that we talked about apply to that area.

Rick Abramson: Um, So yeah, that’s like kind of like the 12 to 18 month. Then you kind of get into like the past the 18 months. And then that’s where, again, things start getting a little bit vague. And we start talking about multi omic computational insights. We talk about predictive analytics. We talk about personalized medicine.

Rick Abramson: Um, I don’t think anybody has really kind of figured out yet exactly what that looks like. Um, and I’m, I’m actually curious to know what that looks like myself, and I’m actually curious to know, you know, I think, I do think that there is a lot of hand waving going on about, you know, AI for, um, you know, for personalized medicine, [00:59:00] and I don’t know what it looks like yet.

Rick Abramson: I’m really, I’ll be very excited to kind of hear a very, very concrete vision for what that looks like, because I haven’t, I haven’t heard it yet. Um, yeah. I was actually just in a conversation yesterday, um, with, uh, with someone, we were talking about primary care. Um, and we’re in a similar conversation to what we were having earlier, like their, you know, private care, uh, you know, primary care roll up, you know, we want to deploy technology, we want to use AI, but, you know, the only thing we’ve really seen in this area so far Is AI for risk adjustment and charge capture and, you know, making sure that we have the right complexity score and we maximize our Medicare Advantage reimbursement, which is important, really, really important.

Rick Abramson: I totally think it’s important. But, um, but it’s not like curl your toes like, wow, this is going to transform care important, you know, and that’s what I’m going to be looking for. And as soon as I find it, I’m going to, you know, Jump in with, you know, I’m going to, I’m going to dive in, you know, cause that, that, that will be, [01:00:00] that will be really exciting once we can kind of, uh, really articulate what that looks like.

Vic: And do you think Europe is the place where these innovations are being tested and trialed or is it, um, all over the globe? What, what do you think, uh, where’s the hotspot that we should be watching for?

Rick Abramson: Well, it’s really all over the place. And, um, you know, a friend of mine actually wrote a really, really provocative, uh, commentary article a couple of years ago, uh, where he suggested that the, as opposed to most areas in healthcare where, where innovation has happened in the rich, developed countries, and then kind of secondarily, you kind of diffused down to the underdeveloped and underserved countries.

Rick Abramson: It was actually suggesting that it may actually be the opposite in this case. That it actually may be the underserved areas of the globe that kind of receive the technology first because there is such a shortage of physicians because there is such a resource constraint because technology can be deployed so much [01:01:00] like easier, more easily, more quickly in underserved areas, and maybe that becomes the laboratory and testing ground.

Rick Abramson: And then from there, you know, kind of following the reverse innovation funnel, you know, like the most successful applications kind of become adopted in the, In the in the more developed countries. I don’t know, but I thought that was a very interesting kind of perspective on it, which I hadn’t I hadn’t heard before.

Vic: Yeah, I like I think that’s a good thesis. I mean, especially the autonomous version of a I if you go to a country like India or Africa or so where they don’t have a lot of providers today and the option is give no treatment or offer this autonomous AI, which is Arguably better quality than human doctors, but it’s at least, you know, comparable.

Vic: Then you can start to get decent volume of how is it performing and then introduce it in the right place in the developed world where there’s a lot of doctors and We had the luxury of [01:02:00] waiting to see.

Marcus: Yeah. Yeah. I mean, it’s just like the proliferation of mobile phones.

Vic: Yeah. In the developed world,

Marcus: right?

Marcus: I mean, it makes sense where there’s a lack of infrastructure, but an appetite for innovation, you know, and low regulatory barriers, uh, you know, then the technology can advance appropriately.

Vic: Yeah. And then we, we watch that and, and grab whatever’s really working. Yep. Yep. Yep. So, um, okay, Rick, I want, I want to close with a couple months ago in March, you deliver the plenary address at the European Congress of Radiology, which is a fancy name, um, and title radiology AI and the analog digital frontier.

Vic: So it’s a 20 minute presentation. We’ll link to it in the show notes. I think a lot of our listeners should be interested. in hearing the whole talk and you sort of co present it with an AI entity, uh, which is pretty interesting. But I want to have you focus on, you end that, that talk with six recommendations for [01:03:00] the radiology profession.

Vic: And I wanted to see if you could talk through that with us, and then I think there’s, there’s some, uh, transition to all, all healthcare professionals or investors or entrepreneurs could take pieces of these recommendations for themselves. So maybe let’s talk through those six ideas that you closed with at the Congress.

Rick Abramson: Yeah, sure. Um, See if I can remember them all.

Vic: I have them written down so I can, uh, I can prompt you if you want, but the first was really define the problems that are need of solutions when we combine humans and AI, that that’s sort of where it started.

Rick Abramson: Yeah. Yeah, exactly. Um, yeah, this was a fun talk to give.

Rick Abramson: It’s, I think it’s out on YouTube. You said you’re going to link to it. Yeah, I’ll link to

Vic: it. It is on YouTube. I think

Rick Abramson: it’s the, uh, the first and only time that I’ll interact with a 20 foot animated AI robot. Bye. So anyway, um, yeah, so, uh, the, the, yeah, the first one was defining the problems in need of solutions.

Rick Abramson: And that, that was really, uh, it was, it was, it was [01:04:00] really kind of like a recapitulation of what we talked about with the academic medical centers in this, this kind of history of, um, of some false starts, I think, in terms of uh, applications being developed that. That don’t really hit the mark. At least like it are really hitting the nail, like squarely on the head in terms of what’s needed.

Rick Abramson: Um, and so, yeah, there’s a lot of developers out there and this is kind of like to the folks out there who are the entrepreneurs and the developers in terms of, you know, when they’re, when they’re, when they’re trying to launch new ventures, you know, To make sure that you get, uh, input from the end users, make sure that you’re in court, you know, that you’re talking with, uh, you know, if they’re clinicians, if they’re administrators, you know, whoever the end user is going to, going to be, um, you know, make sure that you, you do the, the upfront research and, and, and, and help them ideally incorporate these folks in the design process, um, so that you’re actually, you know, defining a solution squarely in need of a, of a, of a pressing of pressing problem.

Rick Abramson: I think the second one was about, um, understanding, um, [01:05:00] how AI tools, um, should be trained. Um, and again, kind of working with developers to, um, you know, there’s different approaches to, to training these tools. There’s different approaches to, uh, gathering data. Uh, you want to make sure that the data, you know, like it is is appropriately reflective of the population.

Rick Abramson: You’re going to serve. Um, that’s a big pitfall. I think that that a lot of developers miss, um, as well as knowing whether are you going to have supervised learning, unsupervised learning. Are you going to do an LLM approach? Are you going to do a more traditional kind of data science machine learning approach?

Rick Abramson: Um, All of which is really part of that whole initial design process and making sure that you’re really developing a solution with the end use case in mind.

Vic: Yeah, and sorry to interrupt, but the healthcare industry, I think you were speaking to radiologists, but the healthcare industry broadly needs to learn enough about AI to then contribute to how it should be trained, and then also you were talking about guardrails in the talk.[01:06:00]

Vic: Where should the guardrails be put, put in place, the clinicians and the healthcare administrators and even the regulators? should learn enough to be able to have an intelligent conversation with a technologist in that space.

Rick Abramson: Exactly right. And a lot of people need to be involved in this conversation. I mean, not to, not to, um, uh, you know, get off on a tangent, but, um, you know, we all know that all, I mean, we have concerns.

Rick Abramson: about, uh, about lack of transparency. We have concerns about privacy. Uh, we have concerns about bias, uh, and, uh, in drift in these algorithms. Uh, we have medical legal concerns. We got to get the lawyers involved at some point to help us understand, you know, what are the implications if all of a sudden you have an autonomous, you know, chest x ray reader, you know, what if, if there’s a mistake, who are you going to sue?

Rick Abramson: Like all these things, you know, these folks, right, need to be involved in the, in the, in the conversation.

Vic: Yeah, then three, three and four were work with regulators and policy makers. And then four was create independent data sets for testing evaluation, which I think is a [01:07:00] really valuable thing that, um, the industry should do so that we bring the right, the right data points in.

Rick Abramson: Exactly. And I think the conversation with policy makers and regulators is all about, you know, walking that fine balance and making sure that we’re keeping patients safe while also promoting access to cutting edge care. And it’s, it’s, we, you know, regulators, that’s the dance that they, that they do. It’s, they’re always walking that fine line.

Rick Abramson: Uh, but, um, you know, we want to make sure that we’re incentivizing innovation. We want to make sure that we’re incentivizing development. Um, But we don’t want to go so fast that we miss the boat in terms of guaranteeing safety.

Vic: The next was probably the most intriguing. Evolve our vocabulary to focus on long term patient outcomes, which suggests to me that people in the audience maybe have a vocabulary that Is not focused on that, [01:08:00] which I was surprised about, but

Rick Abramson: I think so. This, I think, as a problem is pervasive all throughout health care.

Rick Abramson: And this is a, uh, you know, this is a soapbox that I could jump up and down on, you know, but, um, you know, we all, especially in the scientific community, especially in the clinical science community, um, We have a tendency to adopt, um, surrogate measures of performance, surrogate measures of success, uh, because they’re easy.

Rick Abramson: Um, in the AI world, it might be accuracy. For example, you know, we’re gonna, I’m gonna do a study and I’m gonna show that this tool is accurate for detecting, I don’t know, fractures or something. Um, and that’s great. That’s wonderful. But it’s one thing to show that I’m accurate for fractures. It’s another thing to show that the fact that I’m picking up the fracture actually is important long term downstream to the health of the patient.

Rick Abramson: You know, it could be that that’s actually just a tree falling in the forest and, you know, nobody’s really around to hear it. Maybe, you know, it could be [01:09:00] that I’m picking up on fractures that are so easy that I’m not making any kind of impact on clinical care whatsoever. Or it could be that I’m picking up on fractures that.

Rick Abramson: really aren’t important because they’re going to heal by themselves. So it’s only if we talk about long term downstream patient outcomes that we are, can, we can calibrate ourselves to make sure that we’re actually talking about things that are important. And that’s true not only in radiology, it’s every clinical discipline.

Rick Abramson: Um, it’s clinical, it’s non clinical. Um, we want to try to focus on, on, on, On what’s what matters to patient at the very, very end and end of the day. Um, and we could again, talk for hours and days and weeks about this. Um, but, um, but I know you’re familiar with it. It’s important for investors as well. You know, we want to talk about not what’s important to the next quarter, but potentially what’s important to the life and health of a company over the course of multiple, multiple years.

Rick Abramson: It’s the same mentality and it’s just applied very narrowly in this case to AI.

Vic: And then you ended, and I think it’s a good way to end [01:10:00] our discussion. The last one is shape the role of the human radiologist to use AI and other valuable technology tools as the ultimate information integrator of the diagnostic ecosystem.

Vic: So like basically the radiologist would be Um, integrating information from all types of tools, inclusive of AI. Uh, that’s a pretty interesting role to, for the radiologist to become, um, I think, unless maybe I misunderstood what you were saying.

Rick Abramson: Um, yeah. Well, I mean, and it is a great place to end because I think it’s, you know, as we look to the future.

Rick Abramson: Um, you know, we, we all are going to have to redefine our own identities, um, as professionals. And, um, you know, that’s true for healthcare. It’s true for investors. It’s true. I mean, it’s true all over the place, you know, with technology, our, our roles change because technology is, uh, in some places replacing what we do in other places, it’s [01:11:00] augmenting what we do.

Rick Abramson: Uh, in other places, it may make what we do completely irrelevant. So, in radiology, I think, uh, like the professional role, the professional identity of a radiologist is going to be quite different in 10 years than it is today. Um, and I think radiology as a specialty needs to be comfortable with that and needs to, uh, embrace it and needs to get out on top of it.

Rick Abramson: Um, in turn instead of bearing its head, uh, in, in the sand. Right? Um, and so, you know, if you take a look, uh, and I’ll, and I’ll just draw a historical example. So, um, my, um, uh, my father was a pathologist, so he, he, he worked in a, in a patho, worked in a hospital pathologist, um, over the course of his lifetime.

Rick Abramson: The identity, the role of a clinical pathologist was dramatically changed. It was nothing toward the end of his career. Like it was at the beginning. Um, you look at clinical pathology, you know, your clinical pathology, you’re doing, uh, blood chemistries, you’re looking at a microscope, you’re doing blood analysis, all this stuff, the stuff that was [01:12:00] done by hand, by the pathologist, when my, when my dad was starting out doing this and by the end of his career, you know, it’s machines and technologists doing it, and it’s not that the pathologist.

Rick Abramson: had no role there, but it was a different role. Like Nan, you know, the role is supervising, you know, calibrating the machines, performing quality assurance, making sure that the technologists are doing their job right, um, is a totally different role. Equally as important, maybe even more important, but, but very, very different.

Rick Abramson: So, you know, translate that over to radiology. Maybe in 10 years, the radiologist role, if we have an autonomous chest x ray reading machine, for example, you know, maybe, maybe the radiologist role is actually supervising and calibrating that machine and performing quality assurance and making sure that the communication protocols are set with the other physicians and making sure that the work workflows are optimized.

Rick Abramson: Very, very different role, um, requiring different skills, um, but still very, very important. And still, I think I would argue requiring. You know, medical license, postgraduate training. It’s [01:13:00] not like all that stuff is. That stuff isn’t irrelevant. It’s just you’re taking that same training and skills and you’re, and you’re applying it in a very different way, taking advantage of a technology and enabled environment.

Rick Abramson: Um, and so that’s kind of what I was getting there, too. And I think that’s, that’s true for all of us, you know, in medicine and elsewhere, being comfortable, adopting, adapting to new technology and making sure that we can kind of redefine ourselves to make sure that we’re maximally relevant in this brave new world.

Vic: Yeah, so I, I really liked it because I think it’s optimistic, but also pretty honest about what the role’s gonna need to evolve into, and it’s interesting to think about a radiologist, because over your career, I’m sure there’s been new imaging, um, modalities that have come out, or different uses of those modalities, so PET scan versus CT versus MRI versus x ray, there’s probably others.

Vic: And your role becomes helping the physician and the patient decide, which should [01:14:00] we use? How do we interpret these? What’s the right use case? How do we calibrate the machines? What, what slice should we take? All those things are as important as the reading of the particular image. And it’s a, it’s an evolution that I think, uh, is framed really optimistically for radiologists and I think we can translate it to other domains.

Vic: They’ll have to use lots of tools, including AI and other technologies, but the, the medical expert, the doctor, is still going to need to be the, you know, the empathetic caregiver that’s sort of advising the patient on which tool to use in what, in what form and for what purpose and with that whole long term patient care as the, uh, the overarching goal.

Rick Abramson: Absolutely. A hundred percent. Couldn’t agree more. And I’ll, I won’t go into that part where that research show that chat GPT is more empathetic than doctors. We won’t, we won’t mention it.

Vic: Yeah. I think once we free up doctor’s time, They will [01:15:00] become very empathetic. That’s my optimistic. There we go.

Rick Abramson: I like it.

Rick Abramson: I like the optimism.

Vic: Yeah. Rick, anything else

Rick Abramson: that we didn’t cover that we should have? Oh, this has been great. No, I think this has been a great conversation. I really enjoyed it. Thanks so much for having me on. It’s been a pleasure.

Vic: Yeah. Excellent. Well, thanks for your time. Really appreciate it. And we’ll keep following this.

Vic: We may have you back as we get new progress.

Rick Abramson: Thanks, guys. It might be a robot next time, but okay.

Vic: Thanks, Rick.