Show notes
Scot Morris, OD, and Rehan Ahmed, MD, on a Silicon Valley AI startup that bought a radiology practice instead of selling software to it.
The hosts look at what the move from subscription to ownership means for image-heavy specialties like eye care, and how AI-native practices could overhaul clinical workflows.
They explain how machine learning can find clinical patterns at the pixel level that people aren’t built to see.
Finally, they unpack the line “AI won’t replace doctors, but doctors who use AI will replace doctors who don’t,” and argue that the key is using technology to solve specific practice pain points rather than chasing shiny objects.
Transcript
Welcome to Real Talk on AI in Eye Care with your host, Dr. Scot Morris. And I'm co-host, Dr. Rehan Ahmed. In each episode, we discuss current news, add in a little AI education, and debate innovative AI tools and topics that will change the industry. All right, Scott, for some news of the week. This is an interesting one. I'm really curious your thoughts. So this is a story out of Silicon Valley that I think is actually bigger than it looks. So it's two people. It's a product lead and an engineer. It's a Y Combinator startup. They have a background in healthcare, and they started a company called Radly. And it's an imaging company, sort of an AI-based company designed to help radiology. Here's the interesting thing.
They actually, this past week, they didn't sell their tool to radiology practice. They bought a radiology practice. They bought the radiology practice. They purchased. They bought a radiology practice, kind of like a tech company acting like a private equity company. Well, that's flipping the model on its head. Exactly. And sort of makes sense for radiology when you think about it. So here's the value profile for Radly if you go to their website. And radiology is really, I would say, even ahead of ophthalmology and all the imaging AI. So an AI agent has already drafted the report in under a couple of minutes. The radiologist reviews it, corrects it by voice, signs it.
Then you have an AI do follow-up coordination, claim submission, insurance, verification, all those things. The physician still makes the final call, of course, but nearly everything surrounding what the practice and the physician do. Drafting, routing, chasing of all the billing issues is handled by the AI. And by the company's account, they're doing a few million dollars in annualized revenue. And it appears that just a couple of people on the payroll. So here's a question, right? Why buy a practice instead of selling software to it? So, you know, in eye care we have, you and I both, we get a lot of... That's really changing from a subscription model to just a purchase model.
Yeah, exactly. I mean, that's really what it comes down to, right? Is they're saying, well, instead of you buying a per-use subscription fee, we're just going to purchase it and then we own the click fees. Exactly. Think about all the companies that sort of like talk to us about AI tools that they're... And they're really great companies. They're promising to increase our revenue or decrease our cost by improving workflow. And that's the same, you know, that's a common story, right? They build a tool, you capture a sliver of the workflow. If you're the AI company, it's good business, right? But really, if you think about it, that's not where the margin lives.
If you really believe, if you really are improving a practice on a global scale that much, why sell them software? Why not capitalize on the income stream? Why not own the entire P&L? Just buy, just purchase all the margin that you're already supplying them with in terms of the improvements in workflow, right? Wow, that's a unique model. So here's the interesting thing, okay? So when I first read this, I saw this, like, you know, radiology, that makes sense. Ophthalmology, eye care, optometry, maybe not so much. But hey, you know, what have we been talking about over the past couple of years of us doing this podcast, talking to other innovators?
Ophthalmology, eye care, optometry, we're also very image heavy. Oh, very, very data heavy. I think after, and look, we are the first one with an FDA-cleared autonomous code, CPT code for retinal imagery with 9229, the 92299. 2229, there's an extra two hours, I forget about that. For diabetic rhinopathy, right? And okay, people say, look, eye care is different. You got to bring the patient in, you got to do slant lab, you have to do imaging. Oh, maybe not very much longer, though. Exactly. With a ton of slant labs, that's going to change. Every week we're talking about a new technology, the robotic cataract surgery, the automated slant lab, the diabetic photographer, these reading centers that we've talked about.
And we've talked a lot about the unbundled eye exam. Everything is being sort of disassembled. And that, to me, is a perfect recipe for, well, some might not say perfect, maybe I should use my words more carefully, but a recipe for this type of thing. I'm super excited. And you heard me last week talk about an AI-native practice. And the thing about electricity, which I've said many times before, but the big gains in electricity, when the invention of electricity happened, were not right away. It wasn't retrofitting electricity onto a factory. That wasn't the efficiency. The efficiency gains only happened when you imagine the factory having electricity all the time.
Every single process had electricity. It changes your entire workflow. And it changes your entire workflow. I think we're getting in the verge of changing the entire workflow. And I find it interesting that companies buying. I mean, that's putting your money where your mouth is and saying, not only do we believe our technology is great, we believe it's going to change the way radiology functions. That's a big, powerful put your money where your mouth is kind of statement. Yeah. So, you know, the hot take from this week is AI is never the product. Or maybe I should say AI was never the product. The practice is the product. The care that we deliver is the product and the service.
So I think that's definitely a little provocative there. I'm really excited to hear our listeners' feedback on that. So, Scott, I know you have a really interesting knowledge about it for us. What are you going to be talking about? Well, you know, we've been talking so much the last couple of months about data. Data this, data that, you know, data's new oil, data, data, data. And we keep coming back to that fundamental problem. You know, I mean, think about this. Every single day, each one of our clients produced just, I mean, mountains of day with OCT scans, visual field tests and fundus photos and EHR notes and refractions and visual acuity.
And I mean, there's just so much data. And I don't think we're even doing a very good job of collecting it, all of it. But I mean, even what we do collect, there's a lot of data. But here's the point you and I brought forward a couple different times around. Having data is not the same as having insight or knowledge. Data that's not acted upon is just data. Data that's acted upon, that's insight. That's knowledge. That changes things. So I always think about this. We have thousands of patient files, which most of us do. How do we use that data to spot disease progressions? Like, you know, I mean, how do we tell who's at risk for glaucoma? Who's at risk for diabetic rhaba?
Who's at risk for care to go? Who's at risk for myopic progression? Who's at risk for getting a cataract early? And this, I mean, what we're searching for is, candidly, when we're searching through that, millions and millions and millions of points of data, we're searching for needle haystack. And this is where the subject for the knowledge body of the day comes in. It's called data mining. And data mining is very simply put, the process of analyzing massive sets of raw data to discover hidden patterns, anomalies, and other correlations. And, you know, though I'd love to say our gray box, right, we've talked about that before, the gray box that we called our brain, that we could do that.
But the reality is our brain was never designed to analyze massive data sets. That's just not the way it was built. And some of these AI mines, boxes, are built for pattern recognition much more so than we are. And so you've got to think of, and you sum this up all the time, Rehan, is data mining. Data's the new oil, but data mining is the process of getting the oil out of the ground. And, you know, it really is, though data's the fuel, the engine is how we mine it itself. And AI models can't make these predictions. You know, I've talked about this before, time and time again. We're hitting a critical point where AI models can't get any better because they don't have any new data unless they create data, and that creates its own biases, as we discussed before.
So data mining extracts and organizes all these clinical patterns and allows machine learning algorithms, not necessarily LLMs, different subject, but the old school, I always love when I say old school AI, allowing machine learning algorithms to train and adapt and predict and detect and all that kind of stuff. And so, you know, from a clinician standpoint, I know this can sound really abstract, like, gosh, how's it going to do that? Well, and I think that that's where things are starting to change, is that we're starting to get with much of the, let's take OCTs, for example, we're having AI that can scan, you know, millions or tens of thousands of little data points and see little tiny defects at the pixel level that we can't see.
Well, data mining is going to say yes, but this pixel pattern keeps showing up time and time and time again, and maybe we need to start looking for this pixel pattern in all of our OCTs and see, can we predict when things are going to happen? Because our brains, we just can't do that. And so, you know, what's the impact going to be? I think it really comes down to three things, is that, you know, algorithms live by that rule. Our friend Dina always says, you know, garbage in, garbage out. And so data quality is really important. So data mining, you always have to monitor what the quality is. We have to standardize devices and systems, and there's where I think we, you know, I think that imaging devices have, you know, we have DICOM and some of the other standards.
I think that's great. Our problem is EHRs still don't speak the same language and can't talk to each other. We beat that horse every single week, but nothing's changing. And then last but not least is, you know, we always have to worry about privacy and security. So I think that when we think about data mining is, there's still challenges, like is the data that we're putting into the system, maybe not necessarily from a OCT or a retinal image, but is the data that we're putting into an EHR quality data, right? Because if it's not quality data, it doesn't make much sense to mine. In fact, we may actually create biased problems by providing bad data.
Yeah. You know, so I mean, I think that as we start getting into, you're going to, the audience, you guys are going to start hearing that term data mining, data mining, data mining, and it's, you know, and kind of building on what we were just talking about in your news of the day, when we start getting into these AI native practices, you know, the next evolution, where we truly are capturing quality, intensive amounts of data. You know, I think our data mining is going to get a lot better and that's going to feed into our machine learning, which is going to get a lot better. But I think, you know, we need an evolutionary change in how we collect data, how we save data, how we retrieve data, and how we process data.
Yeah. I agree more. Yeah. And as eye care providers, we sit in a very privileged position where we are inputting tons of data. It kind of links back to the news of the week, the values in the data, the values in ordering the imaging and interpreting the imaging. That's incredible value. And that's how the AI systems were trained. And so, you know, as eye care providers, we sit in a very special place. And, yeah, really, really like the data mining knowledge by Scott. This week on the fireside chat, I want to talk about, and Scott, I would really love to get your thoughts here. You know, we often say this line. I know you do, too, when you're in audiences and someone will ask kind of to raise their hand or kind of really nervous to say, hey, is AI going to take my job?
You know, are we going to be replaced? And the classic line, and we've said it on this podcast, I think, it's, hey, you know, AI is not going to take your job. But doctors who use AI are going to replace doctors who don't use AI. Absolutely. And, you know, I did that. And that's definitely a, on the face of it, a true statement. But I want to spend a little bit of time this week on talking a little bit about the nuance of that. Because I think it's important. And, you know, there's some subtlety and deeper meanings here. And this is really courtesy of our talk that we had with Carl Speer, you know, back in Italy, we were on the train, and he talked about this.
You know, what's, what he knows to be true, and he's not totally even drawn the AI comment, but what he knows is that people who, doctors who put the patient first, who don't buy tools for the sake of just a shiny new object, but who are willing to adopt new technologies, AI included, for the good of the patient, that will be good for the practice. And those other, those are the doctors who are going to succeed. And this new sort of medicine 3.0 type of world. And we definitely agree with that. By no means are we saying that you just bolt on AI into your workflow and say, well, now I'm using AI. I'm going to replace the doctor who doesn't use AI.
That's not what we're saying at all. And that's not the way it works. Yeah, Rehan, I so agree on that. I hear people go, what piece of AI should I buy to, you know, to bolster my practice? And I said, I don't know. I don't know. I mean, where's your workflow problem? What is your challenge? Exactly. You have to take the time to sort of look at what is not working for me right now. Yeah. And I think that, you know, I love the word when you say nuance of AI is not going to replace this, but the doctors AI will. I think it's not because they're using AI. It's because they're figuring out how to solve the obstacles and challenges of the current day medical system, which is, you know, I mean, and for any of you who are listening, if I'm going to pitch off with this statement, I.
Okay, fine. Our healthcare system is so freaking broke. I mean, it's so broken. And, and how do we deliver? How do we go back to delivering amazing care in a way that's going to be, you know, profitable for us, value savings for, for patients, improve efficacy, that kind of stuff. And we have to change the way the rules around us are changing and we need to adapt and evolve. And I think the part of that is, is that certain tools, as some of which are AI, some of her, something different, you know, we, we have to start embracing those tools and the doctors who are on those providers, practices, business, people, technicians, opticians, ophthalmology, optometrists who are on that innovative curve saying, you know what, I'm going to wrap my arms around something cause I see this pain point and I want to find a technology that's going to fix this pain point.
And once they achieve that, then they say, well, that wasn't so bad. I'm going to do the next one. Those are the practices that are going to change and threaten the practice to say, I'm just going to keep doing it. And I'm going to race myself to the bottom to see how, how many patients I can see at a zero profit margin. You know, I think that it's about attitude and attitude, you know, it's that old phrase of your attitude determines your altitude. And I think that AI is one of those, the fuel to, to if you have the right attitude, it's going to fuel your altitude. Yeah, definitely. Starting, starting with a specific problem you have in mind from an administrative operational problem to a clinical diagnosis type problem, you can just start somewhere.
And that's what we mean by doctors who use AI. We really are saying doctors who use AI to solve a specific problem and learn about AI and new technologies for the benefit of their practice and the benefit of their patients. Those are the ones who are going to succeed in this sort of new, new wave medicine. I, yeah, I ran. I think well said in summary is that as we repeat pretty much every week is if you're not involved in using AI, it is time to start. Even if it's as simple as getting on Gemini or Claude and doing, you know, asking, how could I write a better letter? Getting comfortable with these technologies is going to be important because if you wait too long, you're going to get left behind.
Yeah, absolutely right. Another really great, great, great learning, Scott, you can find all of our great learnings and our podcasts, our innovators series at aiineyecare.com. That's aiineyecare.com. Feel free to reach out to us at Rehan, R-E-H-A-N at AI in Eye Care.AI or Scott, S-C-O-T, 1-T at AI in Eye Care.AI. Scott, another great episode. Looking forward to talking to you next week. Thank you.