Founder Stories: Cameron Andrews, Founder and CEO of Sirona Medical

Well, everyone, welcome to the Imaging Wire Show. My name is Jake Fishman. I'm the host of the show and the editor of the Imaging Wire newsletter. And we have a great episode for you all today. It's our first one of 2023. And we're celebrating with Cameron Andrews from Serona and Chris Wood from Reveal DX. And we're gonna be talking about a really important topic and one that's gonna be increasingly important in 2023 and every year into the future. And that is value-based care and how radiologists and radiology will thrive in that future environment or in some cases current environment. So we're just gonna jump right into it. Cameron, Chris, welcome to the show.

Thanks for having us.

Thanks for having us, Jake.

Uh, it's really good to have you here. Um, and, uh, to start us off before we get into all the nitty-gritty value-based care stuff, if you can just give us a quick intro to yourselves and your background and your companies and what you do there, and then, then we'll get going.

Yeah, absolutely. I'll start. First of all, Jake, thanks again for having us. Um, big fan of everything that y'all put out, so we're very excited to be here. Um, my name is Cameron Anders. I'm the founder and CEO of Sirona Medical. At Sirona, we focus on building all of the software that radiologists need in order to practice. So viewing, PACS archiving, speech-to-text reporting, Universal Worklist, et cetera. Build this as one unified cloud-native platform. And by re-architecting radiology IT from the ground up in this way, we're able to build new AI-powered applications that help to make radiologists' lives simpler and help make them more efficient. And like what we're building in partnership with Reveal that help to fundamentally amplify the impact that radiology can have on downstream care and its value by extension.

So hi, I'm, I'm Chris Wood. I'm a medical physicist by training, but I've spent my entire career in radiology software. I've started a few companies. Most recently I've been working on Reveal DX, which is a company focusing on triaging lung nodules from chest CT scans.

One of the very first stories ever in the Imaging Wire was about, uh, Qlairia. So it's been interesting to see I've been following you probably longer than, than you've been following along with ImagenArc, but it's been cool to see. Um, so, so Cameron, maybe let's start off kind of with that value-based care future, and, you know, we'll go just start with that goal. So like, what is it going to take for radiology to truly thrive when healthcare is about, uh, operating in just a full value-based care?

Yeah, it's, it's the most important question. I will go into detail, but I'll start with the punchline, which is Value-based care will either usher in a golden or a dark age for radiology as a specialty. And radiologists are really in the driver's seat as to which of those paths end up materializing. So before I go into detail, if there's one thing that I hope people take away from my answer, it's the following. If radiologists demand that the software foundation of the practice, radiology IT generally, is re-architected in order to power that future, then radiologists will not just own the value of all, I mean, sort of image analysis in medicine moving forward, but as we'll talk about as we get into the section talking about what Reveal's doing and what we're building in partnership, own an amplified version of that. But if they don't, the value of image analysis will be systematically siphoned off and radiology's position in medicine will erode as opposed to being augmented and amplified. So from a fundamental perspective, you might say that's alarmist, but AI companies are selling to referring physicians legacy radiology IT vendors are building for other specialties and calling it enterprise imaging. These things should alarm radiologists to the fact that this trend and this nexus, this point of decision is really here today. So with that high-level context, I will now get into details and I'll answer your question. Let me describe what radiology's value-based, why radiology's value-based future really requires this IT rearchitecture. I wanna start by framing the problem in the following way. Radiology doesn't have a value problem. It has a value capture problem. For the last 20 years, radiology has gotten better, it's gotten more valuable, and has gotten paid less as a result of that. And so, you know, one of the perks of this job is you have the opportunity to hear people talk off the record about things that matter. And I've heard a C-level executive at one of the largest payers in this country say, as a direct quote, radiology is the most valuable specialty in medicine and we will never pay them for it. And so the punchline there being what radiologists feel every single day, that radiology is seen incredibly valuable. And in order for value-based care to work for radiology, almost by definition, radiologists have to share in that value. We have to solve that value capture problem first. So the great news is technology can help with that. And that's part of why Chris and I are here. And so the magic formula as to how we get there, maybe there isn't one, but here's one that I'll present. Which is that instead of focusing on how we can get radiology paid for the value that it creates at a radiologist level, at a practice level, so on, which is incredibly important but hasn't really, hasn't really worked very well up until this point. If instead we approach the problem of quantifying and demonstrating radiology's value like a medical device company would, then we can start to make some really good progress. So 3 steps that we think are important for that, for this to work. The first of which is if you, if you mentally decompose radiology into its component subtasks, don't think about reading a chest CT and how we get radiologists paid more for that, but breaking it down into its individual subtasks like characterizing a lung nodule more effectively as either malignant or benign. If we can decompose radiology to that level, we can then use AI to wrap that decision up, make the decision, help radiologists to make the decision more effectively But then critically demonstrate and validate and quantify the downstream impact on the cost and quality of care. Cost, of course, coming first, because that's what allows payers to meaningfully agree to pay for such things. So you take those algorithms that help to improve the radiologist process at very specific points in the decision process, prove the downstream savings as a result, and then if you can bake those capabilities into the radiologist's primary workflow, Now you have something that allows radiologists to keep up with their, you know, with the case volume, which is only gonna get more dramatic over time and something that makes them more valuable in turn. And if you can do more work and be more valuable in the process, then there's value. So in short, what we think is necessary to get there requires that radiology's IT foundation do a bunch of stuff that it wasn't built to do. So in short, realizing radiology's value-based golden age, so to speak, requires that radiology IT be re-architected from the ground up to allow reimbursable AI applications like what Reveal has built and what we're building in partnership with them that can help radiologists capture the value of improving not just one clinical subtask, but over time, many or most of them, and to demonstrate that value to payers in real time.

So Chris, if we just dive a little bit deeper into that AI component and how it can amplify radiology's value in value-based care, can you just give us a little deeper look into that?

Yeah, we're seeing a trend now that AI companies are starting to get reimbursed. We've seen HeartFlow, Perceptum, Optellum, clearly all in that CADx category. And the reason they're getting reimbursed is because there's downstream savings associated with sort of using that technology and having a radiologist use that to produce better outcomes. In the case of the lung nodules, For example, you know, the United States spends billions and billions of dollars a year chasing lung nodules that are benign. You know, our clinical study shows that we can reduce the false positives by about 30%, right? So a radiologist armed with a tool like ours can have this huge impact downstream by reducing unnecessary procedures. At the same time, AI has the ability to accelerate diagnosis. So you can do things like, in our case, find lung cancer early. Our clinical study showed that you could catch about half of the cancers without an interval scan for one of our cohorts. This is really important stuff because when you're rolling out something like lung cancer screening in the United States, you have the possibility of increasing costs. Complication rate for procedures associated with lung cancer screening is about 22%. You can also introduce some morbidity or even mortality associated with these false positives. So AI is really critical to make sure that performance of the radiologist is as high as possible, right? And it's important that radiologists own that tech because if they give it to pulmonology or if they give it to, you know, somebody else, cardiology, then they don't benefit from the reimbursement that's being created around these technologies, and they also don't add as much value.

And the last point I think Chris made is absolutely critical for the following reason: this isn't the first time in the history of radiology where a new technology has come along that has appeared to be valuable and clear enough that it was suggested to be passed to one of the referring— to another specialty downstream. I mean, MR folks were like, these images are so clear, why in the world would orthopedic surgeons not just read the exams themselves? And of course, what radiologists did in that moment was what ended up being, without any question and without any statistical hesitation, best for patient care, which was drive subspecialization. Yes, it is true that referring, that orthopedic surgeons can read X-rays and MRIs, but it is also true that if all you read all day every day are MSK MRIs, you are going to be better than anybody else in the world at doing so. And what radiology has always done effectively is serve as that specialized, then subspecialized, and now hyper subspecialized image analysis experts. We believe that's what's in the best interest of patient care, and we believe AI is the next evolution, in that process of subspecialization.

Hyper subspecialized radiologists armed with the best AI tools is really who I would want reading, you know, any of my exams, right?

For sure.

And if, if you're a payer and that radiologist plus AI is helping you catch disease early or helping you avoid costly unnecessary, uh, follow-ups, then, um, then it's, it's valuable to them and it's valuable to the system. So then in order, one thing to get, in order to start achieving that value, one is like AI has to be integrated well into the workflow. And sometimes I think it's happening right now and sometimes I think it's not. And in the minds of a lot of radiologists, I think they would just say not. Tell me more about that, I guess that challenge and how to overcome it.

There's a lot of legacy PACs out there. Healthcare is not known for being the fastest moving segment of the economy for good reason, right? But there's a lot of legacy PACS, there's a lot of old technology stacks, and integrating AI into the workflow is really non-trivial given that environment in which we find ourselves. And marketplaces have pluses and minuses, and they really haven't been able to solve this problem in radiology. Right. Marketplaces have emerged, but the connections and the ability to integrate and the depth of integration has left a lot to be desired. You know, even our product, which requires one or two mouse gestures, right, is really difficult to integrate into some of the PACs out there. And because of the nature of the upgrade cycles for PACs, where upgrades can be scheduled 1, 2 years out, You know, getting your stuff integrated into a PACS and then actually getting it deployed out into the clinical world can take years. And for a startup company like us, that's just way too slow. And, you know, with AI moving as fast as it is, I think it's too slow for the, for the market as well.

I think the big issue we've had is a framing one in the space. We talk about AI integration. It's PACS integration. We've looked at the AI companies and said, your product, we're really in the, in what I would describe as the third generation of AI's, you know, story in the modern business of radiology. I'll describe what those chapters are in a second. But the punchline is we spent the first two generations looking at the AI companies saying, why can't you make it valuable? Why can't you plug it in? Why can't you get it to where it is that it needs to be? And in fairness, that's partially because Silicon Valley got, God bless, you know, the world in which we come from. But there is oftentimes an intellectual superiority about it that makes us, you know, disappointed to be part of that culture in and of itself. So I understand why people would look at the folks saying we're the greatest on earth and say, maybe you're not as great as you think you are. But that being said, the punchline is without fundamentally fixing what AI companies can plug into, you get what we have today. So the 3 generations as we see it are really the following. V1 was as soon as AlexNet in 2012, 2014, allowed for, you know, modern computer vision architecture to perform at a level that could rival humans in certain tasks. You got the wave of algorithms companies, as I would call it. You had— they solved the FDA problem, but they fell short of integrating into the workflow. And that left the V2 companies, which I would describe as triage as a workflow businesses, with one very clear takeaway, which is let's avoid touching PACS and reporting and any of that radiology IT stuff and see what we can build without touching it. And those businesses really built the first clinical value for AI in medical imaging in, you know, in the last 5 to 10 years. And in the process, they built this clinical value playbook. AI is the engine of a workflow that is valuable for reasons more than just what the algorithm itself can do. But what they fell short of was touching the core radiology workflow. They fell short of usability from a radiologist's perspective. And so the question then for us as the V3 generation, which I would characterize Reveal and Sirona both as being members of, is If instead of avoiding radiology IT, you re-architected it, what could AI do for radiologists that it cannot do today that AI companies simply cannot do? Yes, integration can be easier for technical reasons that radiologists shouldn't care about, but that Chris and I do deeply, and as Chris described better than I could. But the second of which is you could use AI to solve problems that otherwise you wouldn't be able to. You can combine an algorithm like what Reveal has done, which will help amplify the decision-making around lung nodule characterization with an algorithm that we have that's FDA clear that helps automatically measure lung nodules and chest CTs with 15 other algorithms from 15 other AI businesses that help create the best user experience. It's not just about how you integrate in AI, it's about how you power the physician to have the best, most efficient, and most effective experience of practicing medicine. And that's not a step— that's not an incremental improvement on the current status quo. It really requires a fundamental step function change in the technology underneath all of this.

Speaking of step function change, so when we were first starting out and you were describing Sirona, you mentioned essentially like all the different components of the imaging IT stack that typically have had different brands and different invoices and different account managers and everything else. That's, that's not something that I hear coming from one company very often. Can you help me understand that a little bit deeper?

Yeah, absolutely. Um, radiologists would never give a colleague a formal second opinion without having the full clinical picture. And if we want AI to play this role, not just in value-based care, but just software to help simplify and amplify radiologists' lives— and, pardon me, simplify their lives, amplify the impact on downstream care— we have to ask ourselves critically, how the heck could you ever trust the output of an analytical system that also didn't have the full picture? It's not just about what's in the images, it's about the full picture. Radiologists intuitively understand that. So if software is going to help radiologists to do a better job in the future, it needs to fundamentally have not just the full data picture, but it needs to make radiologists' lives easier in the process. So by unifying PACS, archiving and viewing and reporter and worklists and all of these things together, we're able to do really 3 things. The first of which is unify the data the radiologists need to practice. The second is we unify all the software that they need to practice. And the result of those two is that we can unify an understanding of what the patient— of sort of the patient, uh, sort of in full. And as a result of that, we can understand the full clinical context as a radiologist is reading line by line, image by image through that case. So if unified is the product value, that this architecture can bring to radiologists, then cloud native and AI, they're the toolkits that allow us to achieve that unification. Who cares if you have all the, you know, report and all the image data in the same place? What matters is what you can do with that information to make radiologists' lives better. Things like solving hanging protocols all the way through to making it possible to click on a certain piece of anatomy and get lab values associated with it. That's the value of unification. It's not just having the data and having the software, it's what you do to bridge those two together using artificial intelligence and cloud-native capabilities that I can go into greater detail on, but are fundamentally different than what exists today.

And you need an understanding of, of value, and this needs to be central to sort of your, your goal, right? Radiologists in our, in our business, you know, triaging lung nodules, um, it's the downstream physicians who have to deal with getting patients to come back if, if the radiologist says you need an interval scan, right? And we know about half of them won't come back, or won't come back on time at least, right? So the value that the radiologist brings is much higher if they can figure out, if they can inform what to do with that patient today and now. You know, just saying come back in 3 months, coming back in 6 months, the value that— that might be fine clinically, But in the real world, what that referring physician is dealing with now is trying to get these patients to come back. And that means we have an opportunity to add value just by, even if we provided the same answer, the same results without that second scan, you'd be adding a ton of value, right? But we actually are able to improve the performance as well as trying to accelerate that decision about what to do with that patient. So that you're not stuck in this cycle of doing repeat scans, which does nothing but just sort of increase cost and really reduces the quality of care because the non-compliance is so high.

Karen, you said something earlier that got me thinking. It's not just because your background, but the cloud native piece. So, you know, over the last— basically at RSNA 2022, I've heard a lot about cloud and some folks were talking about cloud native and some folks were maybe like, tactfully steering that conversation away from cloud native. But I, I'm curious what the awareness is about the difference between the two and also like what is the difference between somebody who might just say cloud and somebody who's truly cloud native and how their performance might be different?

Yeah, that's a great question. So if there's one thing that folks take away from the answer I'm about to give, it's this, which is that the difference between cloud native and cloud-based IT systems is critical to radiology's value-based future for the reason that I'll describe now. So I started talking about this problem. We're talking about what does radiology need to thrive under value-based care? Talked about the sort of existential crossroads that we view radiology as being in at this point. The very nature of existential problems is that they cannot be solved by incremental improvement on what is, otherwise they would not be existential. And if radiology's value capture problem could be solved with today's IT architecture, the IT vendors would've done it over the last 15 years. Good Lord knows there's value in the process of doing so. Now, so solving existent or solving these kinds of existential problems almost always requires a fundamental paradigm shift in order to achieve that. And that paradigm shift is almost always bad for people who are selling stuff today. And that's what we in Silicon Valley land call the innovator's dilemma. It's worse for the existing businesses to build than it is for them to maintain. That results in— analogies can be corny and this one will be, but fundamentally, like to analogize this to another existential crisis, if you look at sort of the climate crisis, generally speaking, saying that radiology is going to be able to get to its value-based future with the shift from on-prem to cloud-based is akin to saying we can solve climate change if everyone drive a Prius. It's like a halfway solution that doesn't get you all the way there. I'll be very specific about why that analogy is precise, even if it is corny. Cloud-based is very much like saying it's a hybrid car. All of the major vendors in the space are doing what makes sense for them to do, which is taking their existing platforms, which are battle-tested and they work in the ways in which they work, and they're putting it in the cloud. They're running it on servers over at Amazon rather than the servers in your basement. That is exactly the same thing as putting a battery in a gas car and calling it a hybrid. There's value there. It's not a gas car, but it isn't going to solve the problem, which it sells itself to be able to do. So cloud-based is like saying a hybrid car, and cloud-native is like saying electric. The value there is dramatically different. And I'll give just one illustration of the difference between those two and why cloud-native will be as important a technological shift for radiology as electric cars are for transportation. One very clear example of this being both unified and cloud native allows Sirona to make it risk-free and cost-free for practices to start their IT transformation journey. Here's what that means. We can deploy as an overlay with no servers deployed to anyone's data centers as an overlay on top of as many different PACS and RIS systems as a practice might have. We can do this typically in about 20 to 40 hours of one IT resource's time on our customer's end. And we can create one universal workflow across all of them that allow us to deploy solutions like Reveal DX and the whole lung suite we're building around Reveal and through partnership with them and partnership with Chris that allows us to deliver these kinds of powerful solutions with no server costs, no installation, and to do so as an overlay from the start. So if radiology IT is a gas car, cloud native or cloud-based is hybrid, cloud native is electric. And the value between those is very substantial and means a lot to radiology's value-based future.

Chris, from a Reveal DX perspective and deploying into a, you know, cloud-native unified platform, how is that different or how does that streamline the AI integration process?

Well, as I mentioned, you know, it can take us quite a long time even with customers asking their PACS vendor for us to actually get deployed into a clinical practice. If, if we have a cloud infrastructure and we're able to participate in sort of the Serona cloud-native system and be embedded in that, as Cameron said, he can deploy this out on top of any PACS, and we now have the ability to deploy into that practice very quickly, that same 20 hours, you know, we're included in that, right? So it provides for us a faster way to market. And it removes some of the things that are hurdles for us, like spinning up servers, you know, on-prem. If we want to sign up a customer, they just sign up. They don't have to provision all this resource, and we don't have to go through a lot of new security checks, you know, every time they bring something inside their firewall and they have to evaluate this new application that's coming inside their firewall. Now, as part of Sirona, we're basically able to access these customers much more efficiently. And as Cameron said, we're on top of this, we're not just embedding inside Sirona, but we're actually building something specifically for chest CT. You know, chest CT is, if not the slowest exam for a radiologist to read, it's one of the slowest exams for them to read. And with lung cancer screening right now sitting at about 5% of the eligible population, The number of chest CTs is gonna go up and up as the number of people getting screened goes up and up. AI is gonna start identifying other people who are nonsmokers who are also in need of lung cancer screening. That's gonna add to the total burden. So you have all these radiologists with this slow exam, you know, that's hard to read, getting inundated with all these new chest CTs to read. They're gonna need something. And we're working with Sirona to build something that's kind of bespoke for chest CT. That will allow them to read these much, much more quickly.

I think to augment everything that Chris just said, here's what radiologists need to know about cloud-based. It's not new software. If I take a piece of software and I run it on my MacBook Pro at home or I run it on my PC across the room, it's the same piece of software. What cloud-based is, is the same PACS software running on your servers that's on-prem versus running on Amazon servers or Google servers or Microsoft servers. That's cloud-based. It's the same systems. And so we don't have to prove that cloud-native is better based off of the fact that we were describing earlier what the issues are of integrating AI with the current workflows. The proof is already in the pudding. It is no easier or no harder to deploy AI to a cloud-based system than it is an on-prem one, because it's the same software, it's the same solution.

Let's look into the future and, um, Value-based care is a reality across all of healthcare, and parts of radiology or specific practices or specific organizations are, are indeed thriving in value-based care. What are the things that they did? What are the actions that they took between now and then to get to that thriving state?

I'll start just quickly. The most important thing for radiology practices to focus on at this point is probably what Southwest Airlines should have done 2 years ago. Which is, uh, to go ahead and start layering stuff on top of a broken foundation is a recipe for disaster. And so instead of saying, how can we layer one algorithm or one thing or one capability on top of the underlying foundation that we have, how can we unify and simplify our technological foundation in a way that allows us to streamline operations such that when applications like what Reveal has built come along, and many others that will be built for other application areas over time, that it is no sweat off of our back to deploy those in turn, to look fundamentally at the fragmentation of the IT systems that folks have, and by whatever means, in whatever way they see prudent, to unify them and by extension then simplify their underlying IT operations. It is impo— there are two kinds of problems in software, one of which is totally intractable, the other of which is quite simple. If you have a problem that scales exponentially with every single IT system you have to integrate it into, meaning one AI algorithm, you got 5 PACS systems, you gotta integrate it 5 times versus one that is a single push of a button and it's available to all radiologists in a given practice immediately. The delta between those two things when you start to get to scale 10, 20, 30 AI algorithms, swapping one out for another one, et cetera, becomes really difficult for practices to manage. You cannot get AI deployment at scale without unifying the data and the software layers that radiologists use to practice. That's what I would focus on doing.

And I would jump in and say that radiology practices now that are embracing sort of their need for analytics, and this is clinical outcomes-based sort of leaning analytics where you're able to find out what impact your report is having downstream. And a little bit more on standardization, you know, practices that are trying to standardize their reports, make them more structured so they can pull analytics out of them. Those groups are going to win. I mean, again, with, you know, the case of lung nodules, because that's been my life lately, you know, there was a study out last year that You know, the variation between recommended additional imaging between one radiologist and another for a chest CT can be a factor of 4x. So here you have one radiologist recommending 4 times as much additional imaging as another radiologist in the same practice. Now, if they had structured reports and they had a way to track that, right, and they understood clinically what the downstream impact of that was, that's something a practice can correct. And then they have to also get good at publicizing that to their— the groups they're reading for, um, so that they can really push out how much value they're adding. And, and, um, once they— once they can prove that, then they're going to thrive in this new value-based care environment.

We believe that today in the radiology world is akin to the EMR market circa 2009. For the following reason, which is, as Chris alluded to and described, CMS and payers are beginning to reimburse materially for AI, broadly speaking, starting with the application and lung nodule characterization capabilities that RevealDX is helping to pioneer. That is the economic incentive that might drive practices to say, hey, well, if there's reimbursement behind what these folks are doing and it helps amplify the value of our care to downstream, value of our practice to downstream care, we should adopt that. And yes, they should. That's great. But the point is this. If you in 2009, or whatever you want to say about the EMR companies that have ended up winning, if you bought the third best EMR in 2012, you were feeling pretty bad. And making no decision may have been better than making the wrong one. And so the point here, I think, as we would describe, is the economic incentive The AI wrapper around radiology points of decision that by making radiologists better at making very specific calls, getting payers to pay for the savings they get as a result, that reality is here now. But the underlying infrastructure necessary to do that for not one but all of those applications over time, that's what we're building. But it's not as a broad category of thing. It doesn't have 10% market penetration today. And so the question that I would ask, or that if I were a radiology practice, is not, What is the solution that gets us to where we want to be 2 years from now? It's does the underlying IT foundation we have support a future, the value-based future, and in closing that value capture gap that we described earlier is critical to the value-based future of the specialty.

That's actually a really, really good point that it's kind of hard to predict the future, but it is a more straightforward task to just make sure that you have the IT infrastructure to be able to act and react. That's really cool. And to both of you, I mean, thanks so much for coming on and sharing your thoughts. And, um, it, you know, uh, this is, uh, I'm speaking with a couple people with some big ideas, and I find that to be, uh, really cool. And the thoughts that you shared, um, and the vision that you shared for how radiology can thrive, uh, makes a lot of sense. So thank you for coming and sharing, and kudos on all the the innovations that you've been working on. Uh, it's been cool, you know, me as the armchair radiology follower, uh, it's been cool to, to, to witness. Um, so, so great job on that, and thanks for coming on. For the folks who are watching, thanks for joining us. Um, if you are, uh, you know, today working on your infrastructure, um, and working on a better way to, uh, measure and characterize lung nodules, um, I think Chris and Cameron gave some really good ideas about how RealDX and Cerner can help today. And then also, I know it's sometimes hard to plan beyond today or this year, especially being the beginning of January, but I hope this kind of helped jog everybody's memory that, that being future ready is of the most importance and maybe gave you some ideas to that. And maybe it gave you a couple people to, to reach out to. Uh, so thank you everyone. Thanks for coming in and, and tuning in, and, uh, uh, have a great 2023, everybody. Thanks, Jake. Thank you.