How AI Is Fueling the Transformation of Healthcare

Michael Marks: Today we're going to be talking about artificial intelligence in medical applications. AI is of course flooding into so many sectors right now, and today we're going to discuss one that has big implications for every one of us. That's healthcare. AI is entering every aspect of the healthcare ecosystem.

From administration to clinical applications and therapeutics and research. 32 billion has been invested into AI healthcare companies since 2021, with another 11 billion estimated in 2024, according to a recent report from Silicon Valley Bank. Today, one in four dollars invested in healthcare companies is going to companies that are leveraging AI.

There's obviously a big opportunity here to lower costs, to address labor shortages, and to improve the health of patients. But it is still early days in this space, and challenges still need to be overcome in driving adoption, understanding the landscape of buyers, regulation hurdles, and other areas. To navigate this topic, we are joined by the CEOs of two emerging AI healthcare companies who are navigating all of this firsthand.

Dr. Alex Sardiña is the CEO of White Rabbit, a company developing products for earlier, more accurate cancer detection using AI. We also have Kalyan Sivasalam, founder and CEO of 5C Network, developer of a suite of AI products for radiology. Alex and Kalyan, thanks for joining us today.

‍Dr. Alex Sardiña: Thanks for having us here, Michael.

‍Kalyan Sivasilam: Pleasure, Michael. Glad to be here.

Michael Marks: Welcome. Okay, so, you are both working at building companies that use AI to improve patient diagnostics. But even within the same vertical, there's a lot of variety and different approaches in business models. Some are focused on early detection, improved accuracy, speed of diagnosis, Expanding access and others.

So let's ground ourselves in where your two businesses are focused today. So Alex, let's start with you. Can you start by telling us about what you do at White Rabbit? What are the problems you're solving in mammography? And who benefits from it?

Dr. Alex Sardiña: Absolutely. So a little bit about White Rabbit. Um, it started a little bit over eight years ago, and the founders had a vision of exactly what we're talking about today is where does a I intersect health care?

But they narrowed that focus even a little bit more on a I and diagnostics. So they started looking at What disease process could be impacted positively by A. I. And they settled on breast cancer and appropriately so As many of you know breast cancer affects one of eight women in the United States But on the flip side of that it's something or a disease process that if caught early has a very, very high cure rate.

So the first reason for the product that we're working on is early detection because early detection in this case does actually save lives. But they went a little bit further. They went ahead and purchased and operated nine imaging facilities to understand additional pain points in the screening process.

And in that process, they had two additional revelations. One of them being that the screening process today is actually fraught with An incredibly large number of unnecessary recalls. And what that means is women are being called back to have additional imaging, additional biopsies, creating a tremendous amount of patient anxiety, but also adding an exceedingly large amount of cost to our healthcare system.

It's, it's estimated that these unnecessary callbacks are costing the healthcare system about $3 billion annually. Wow. So that's a second reason why our product is, is. We hope to be very impactful and an emerging problem since the pandemic has been the shortage of radiologists. There's a complete flipping of the supply demand curve where radiologists are overwhelmed by the number of screening exams that they're having to interpret.

So introducing an A. I. Tool that can actually alleviate, you know, the workflow issues and the staffing issues is another reason. So I think those are the three main reasons and the three main subsets that we're trying to help address with our product.

Michael Marks: Thank you for that. Obviously a very important problem.

Everybody seems to know someone who's touched by cancer of all types. So Kalyan at 5C you have a pretty broad suite of AI enabled products for radiology that are already in the market. You've built a business in India, which has quite different market dynamics. as from the U. S. of course, which we'll get into a bit later.

But can you walk us through the problems you're aiming to solve at 5C?

Kalyan Sivasilam: Well, for sure. Dr. S. already touched upon, uh, you know, the fact of radiologist shortage. Uh, in India it's probably even more acute. Maybe 12 radiologists, of course, for a population of 1. 4 billion, uh, is really not enough.

Michael Marks: What, what's that number in the U.S., just out of curiosity?

Kalyan Sivasilam: It's about 55, 000, 55, 000 in the U. S. and 13, 000 in India and, uh, I mean to add, uh, medical imaging centers in India, like are growing at 16 percent year on year. That's a big growing market because evidence based medicine is really caught up in a country like India. And so there is this radiology shortage, and I'll talk about that a little more.

But there's something more fundamental that's going on, which is why I think a product like 5C is so important for that geography and many other geographies, is that even with the shortage of radiologists, radiologists are specializing. So today, you don't even have general radiologists. There are, you know, people who specialize in intervention, onco, vascular, musculoskeletal, neuro, nephro, cardiac, etc, etc.

Because that's where, that's where the excitement is, and that's where the money is. And we found that the cases, the diagnoses that are most often delayed, most often misdiagnosed. And the cases that actually caused the most amount of radiologists burnout are not the complex cases. That's actually what radiologists are genuinely good at.

And that's what they studied for. It's actually this volume of Mass cases. What Dr. F spoke about, for example, screening, which comes at such scale in the U. S. In India, the scale of chest x rays, you know, for health checkups, etc. It's these basic diagnoses, kidney stones, consolidation, pleural effusion, etc., that come at such a massive scale that it overwhelms the radiology system. And so we believe that, uh, We can build a machine to compliment the radiologist and basically do 80 percent of the work that happens in the radiology department to finally put a, you know, to give you an analogy, I feel like radiologists and doc, maybe you can validate this.

I feel like you guys are a lot like mathematicians, right? You're really good at arithmetic, which is the basic stuff, but then you have graduated to a point where you're doing, you want to do complex work. Some of you want to solve the Riemann hypothesis. Um, If I keep giving you basic arithmetic stuff, you'll be good at it.

But at some point in time, you'll make a mistake. If I give you complex arithmetic a hundred times a day, you're going to start making mistakes. And that's what's happening to radiologists. And instead we're building a calculator for that. It's that simple.

Michael Marks: That's a great analogy. That was a surprising one.

Good for you. Okay. So, there's some skepticism about the amount of AI hype we're seeing everywhere and companies quote, unquote, AI washing.

Michael Marks: Or trying to thread AI into their product strategy after the fact. That's what everybody's doing these days. Mm hmm.

Michael Marks: It's reminiscent of everybody being in the dot com business.

Your two relatively early adopters of the technology haven't been founded eight or nine years ago. Yep. So, I'd like to hear more. In more detail, exactly how as manifesting in your businesses, are you using computer vision, generative AI, deep learning or other forms of AI and explain if you would, why AI is necessary to help overcome the problems you're targeting?

Dr. Alex Sardiña: Absolutely. Look, I think unlike Killian, where actually they're working on a suite of products, we are specifically focused on diagnostics, diagnosing cancer on an imaging in this particular case of mammogram. So AI is at our core. There's no A. I. Washing when you're trying to diagnose cancer, and the only way you do that is by taking massive data sets, teaching them how to interpret different images, teaching them how to discern between normal tissue and cancerous tissue. So you have to have a lot of computational power. You have to have the latest and the most advanced algorithms. So we're using convolutional neural networks and applying those into vision transformer models to help the models and the algorithms be able to discern the two.

There is no way I washing in that you have to have a I at the core of what we're doing. And, yes, there's other unregulated AI that could be used, or companies can be claiming to be using AI, but you cannot be in this space without having it at its core.

Michael Marks: One of the things that I think our, our listeners would, would like to know more about, could you spend a couple cycles and talk about what you do with these data sets?

And I think you have real technical people who are, who are, you know, going through the data. Could you talk about that process a little bit?

Dr. Alex Sardiña: Yeah, absolutely. Again, going back to mammography, why mammography is kind of primed for being kind of like the spearhead for taking AI diagnostics to the next level is because you have a lot of longitudinal data, which is key, right?

You want to make sure that you're feeding these models with the right information. What did the mammogram look like five years ago, four years ago versus last year? Finding those data sets when we first started was almost impossible because health care systems and hospitals hadn't figured out the value of the data had not structured the data in data warehouses.

So purchasing the data or acquiring the data was very difficult. And then you're buying data that you have to curate, which means you have to package it. You have to put it in a format that's digestible by your models. And the other thing that you have to be careful with is protecting patient health information.

For sure. Because we have HIPAA compliance in this country. You have to make sure that all that information has been redacted. So it's not only as simple as getting the data and feeding it into the model. There's a lot of manpower. And I think you can probably speak to this even better. A lot of manpower spent on annotating and curating the data to make sure that the models perform like they're intended to.

It's like the old adage. You get out of it what you put into it. Good data in, great outcomes out. So it's really important that you identify the right data sets and that you have the right people curating it.

Kalyan Sivasilam: And Michael, you use the term AI washing, but I want to maybe not bury the lead and kind of talk about why there are so many companies in the medical imaging AI space.

I think at last count, I at least know 700 companies trying to do medical imaging AI. But I want to try to, again, give another analogy to kind of explain this. Today, there is a clear signal to the market that radiology and AI are a thing, right? There's, there's this radiology shortage. We just spoke about it.

The other thing is that it is the, it is almost trivial to build out a very basic computer vision model. I can bet you can. Right now, in the next 30 minutes, with a MacBook M1, no GPU, the three of us, with Dr. Sardina's help, will be able to build a really, really good classifier for something as complex as fibrosis, right?

It is, it is literally trivial. So the cost to start and just build a basic model, right, is actually zero, Michael. But the analogy that I want to bring here is that, to go to Everest Base Camp, It's trivial. Like you, you three of us can literally take five days off, uh, book a ticket to Kathmandu and then, you know, five days later we will be in Everest base camp.

It is easy to go to base camp. It is very hard to summit, right? And so I like to talk about AI medical imaging AI startups as base campers, but it's trivial to build a basic model and then sit in front of an investor like you and say, I got a model versus people are submitting and the summiting is the is where you need all this data that Dr. Sardina spoke about, where you need to take an insane amount of data, have very clear annotation guidelines, and be able to actually bring models to 99 plus, whether it's a negative predictive value or F1 or whatever your point of accuracy is. That's where I think the difference between all these base campers and the submitters are.

When you look at the number of people who are actually trying to submit. It's very few. The space is not that crowded because data, annotation, clinical knowledge is all very, very, it's very specific. And if you need to summit Everest, you don't wear the same gear that you're going to use to get to base camp.

So you need, you need different gear.

Michael Marks: So where do you get all this data?

Kalyan Sivasilam: It's a great question. We actually had a very specific strategy in terms of collecting data. When we started 5C, we knew that in order to build a great AI company. You need to build a great data company. And so we also realized, as Dr. S just said, that these data sets are just not available. And we knew that we wanted to focus on single shot, Prediction of diseases. And so we didn't necessarily need longitudinal data, but we just needed high volumes of data. And so we actually started our business as an online radiology company, where we would work with hospitals, connect them to a radiologist on our platform, and then store perfectly the images and the associated reports.

And something interesting happened. It is that a few years back, we stopped seeing new diseases on 5C, which literally meant that we have seen, basically, the world of diseases in radiology, which is really exciting. And this meant that we can now map this out and start creating models. So the beauty of our model at five C is that we don't build models on 10, 000 data points like we are in millions.

For example, our algorithm that pretty much autonomously reads the chest x ray today is built on something like three and a half million Uh, chest x ray data points, which are completely annotated and ready to go in.

Michael Marks: And you're getting that data from the radiologists?

Kalyan Sivasilam: Uh, validated by radiologists through reporting, but I'm getting them through hospitals and diagnostic centers in India.

And of course, I must add that, uh, in India also there are clear guidelines today on, on data protection. Yeah. But there are also, but it's called the DPDP, the Data Protection Act there, but it does allow for companies to anonymously use this data. Okay. In order to. technology on top of it. And that is, I think, an India advantage in that sense.

Michael Marks: That's a big advantage. We're going to get to the India advantages in a minute. But Dr. S, you might also, where is your data come from?

Dr. Alex Sardiña: So, we were very, uh, fortunate early on to establish a relationship with a very reputable academic health institution in the Midwest. They actually laid the foundation for the first models that we have.

When I joined the company, I was able to kind of leverage prior relationships and, and bring one of the largest mammography providers to the table. Uh, and they brought a much more diversified. data set, which is another key killing. You can probably speak to that is having not only diversification and disease entities, but diversification, the patient population, diversification in machine manufacturers. Correct?

Kalyan Sivasilam: Sure. Absolutely.

Dr. Alex Sardiña: So, but we've seen, we've seen a transition over the last several years where you're finding more data vendors that have actually realize the value of having good data. So it's becoming more accessible. But that's also driving the prices for the data acquisitions up. So I think your strategy and how you've done it by having a tele radiology foundation where you were able to kind of extract the data and the pathology was ingenious.

Kalyan Sivasilam: And the important part of that was not just the data. It also gives me a reinforcement learning with human feedback loop built into my business. And so I don't need to go out and To collect data. I don't need to go out to enrich it. I build my own models and I do RLHF on my own network and then I can take it to market.

So multiple, multiple benefits by being extremely heavy operationally.

Michael Marks: Perfect. I mean, we hear lots of things about the future role of radiologists. Is AI going to eliminate their jobs? There's also shortages. I mean, is AI going to solve for the shortages problem? So what does that look like from, let's start with you again, Dr. S.

Dr. Alex Sardiña: Well, I can tell you that been, I've been accused of that more times than I. Then I cared her answer.

Michael Marks: Accused of eliminating the jobs.

Dr. Alex Sardiña: Yes. Yes.

Kalyan Sivasilam: Eliminating your own kind.

Dr. Alex Sardiña: Correct. I'm married. I'm married to a breast imager. Fortunately, she's on the tail end of her career and she's like, that's fine. Okay.

Dr. Alex Sardiña: Um, but I, you know, I, I don't think that's going to happen.

I mean, I think the demand and supply curve is so upside down that you need a tool like the ones that Killian and we're moving up. You just don't have enough. And what's happening is. And you know, we all went into, and I'm generalizing here, but we all went into the field of medicine to practice medicine.

And part of that is the physician patient interaction, even in mammography, right? Because when you have to biopsy a patient, when you have to tell a patient, unfortunately that they have breast cancer, that is a very emotional part of your day and is one that you want to spend your time with that patient and give them the quality of your time and education.

Thank you. And that's going away because you can't spend time. You can't spend the 15, 20 minutes with your patient because you're like in the back of your head. It's like, I have, I have a bunch of other cases. So what we're doing is not replacing. What we're doing is making their workflow more efficient, more efficient so they can focus on those cases.

And more importantly on those patients that actually need their attention. To your point, the calculator, right? You want to get back to what you're good at. You want to get back to diagnosing cancer and walking patients through that journey. Where today, that's, that's nonexistent. So, we're not replacing, we're helping them get back to what they intended to do.

‍Kalyan Sivasilam: And one thing that I've seen in India and a few radiologists in the U. S. have also validated, is that, radiologists are not just able to not spend time with patients, they're not even able to spend time with their physicians. And that is a big problem, right? A big part of radiology is time. communication to get to the optimal result for the patient.

And many times, that would involve working with the surgeon, understanding where a particular pathology is, what are the different routes of accessing it, how do they actually take out, let's say, a specific thing. Radiologists just don't have time for such things because Like you said, in the back of their heads, they're just thinking, Oh, I got a bunch of other cases.

Right? And, and at the same time, they're thinking, I need to give this guy, uh, the best of my abilities. And many times, unfortunately, they kind of fall between the stools, right? And, and I think that there's an opportunity for AI to, to genuinely give back time to radiologists to focus more on complex stuff.There's a lot of stuff that AI won't be able to do in the near future, Michael. For example, we talk about AI to do Cancer, and I think breast cancer is a great idea, but there are a lot of cancers that are so that never form a pattern and it's very difficult to never form a pattern pattern and in the end, AI is in the end, AI is pattern recognition.

Of course, I'm simplifying, but in the end, it's pattern recognition. And some of these are so complex that you'll always need deep expertise and deep oversight. And that's what radiologists, uh, And radiologists are the doctor's doctor, right? Head to toe, they understand anatomy better than any single specialist.

And that's why they're so important, that's why they're so good. And building a tool that can take away mind numbing and extremely high volume stuff and give them back time to practice high quality medicine, I think is a, I mean, I would want a radiologist, if they're ever reading a case of my mother or myself or somebody like that, I would want them to have all the time they need.

I would want them to have expertise at their fingertips. And I'd want them to be in a position where they can help me get the best care.

Dr. Alex Sardiña: And today, That's not the case, let's be honest, but I think I think you brought up a good point to is that repetitive kind of mind numbing work that can be offset by an AI tool.

That causes the radiologist to actually, when they're tired, that's when you make mistakes.

Kalyan Sivasilam: That's when they make the mistakes.

Dr. Alex Sardiña: That's when you're going to miss a cancer. That's when you're going to call back a patient unnecessarily. So let's eliminate that. Let's at least make that part of the workflow, something that becomes, you automated or becomes a lot quicker to get through so that you can focus on those complex cases.

‍Kalyan Sivasilam: You know the most misdiagnosed thing in India? What's that? It's actually fatty liver. So a very, a non trivial population in India has fatty liver, but because radiologists are so busy and they're all looking for the appendicitis or the bowel perfusion, that they just, they kind of ballpark and eyeball the size of the liver. In many cases, there is a grade one fatty liver that could definitely do with maybe passive intervention, like, you know, dietary changes, et cetera, which a good report will tell you. And a cool intervention that we have done is no radiologist has to ever measure. a liver on 5C. It's automatically measured with a mean error of like less than the size of the tip of the pointer of the mouse.

So it's always accurate. And so no radiologist has to ever measure a liver. They never have to reconstruct it. They never have to do APCCTR, three dimensional, and so on. It's all done for them and if there's a fatty liver, they will, like we will call out that hey, we think there's a fatty liver and for this patient of this age and this sex, this is the guidance for fatty liver, you can approve it and so it's, now, they don't miss it anymore.

They don't have to like just do this stuff again and again. They can take a decision. Based on an input.

‍Michael Marks: Yeah. So that's fascinating. I, I want to dig in one more layer here because what you hear about all the time is, you know, is artificial intelligence going to replace people, whether it's radiologists or, you know, robots or whatever it's going to be.

But. You, you both touched on a part of this that's really important, which is accuracy and accuracy, as you know, from just reading the traditional press, you know, there's like AI is going to take over the world. And then there's like AI is dumber than my house cat. And so There, there's this whole question about accuracy, and Alex, this is a good one for you because I think our listeners know that mammograms are not very accurate.

There's very high levels of false positives and false negatives, and you were talking earlier about false positives, and people have to come back and do all this work and biopsies and then they don't really have cancer. I've also heard it said that the highest, uh, Level of accuracy for the reading of mammograms is if they're read before 11 in the morning and you talked about being tired So could you give us just a little bit more information about how AI in this field?

We're discussing in the medical field can really improve accuracy and therefore improve outcomes.

Dr. Alex Sardiña: Absolutely, and I think you touched on it, right? There's there's a lot of things that go into to generating a radiology report, whether you're generating a report for mammography, whether you're generating a report for a cardiac CTA or an abdomen, you have to touch on the liver, the pancreas, the kidneys, the adrenals, the bowel, the lymph nodes, the bones, and you have to go through that process again, going back to the remedial work of pointing all of this out.

There's a certain level of pattern exhaustion that starts happening. And that. Impacts your accuracy. You're a lot more accurate as a radiologist. If you can focus on the areas that the A. I tells you to focus on and a screening mammogram, it highlights two or three places. Everything else with a certain level of confidence.

You're like, I don't really need to look at it. Let me focus on these three areas. And then to your point is let me use what I've been trained to do a higher level of cognitive reasoning that A. I is not there yet. That's what drives accuracy.

Michael Marks: That's a great answer. And you really sort of hit the same thing by saying you don't have to look at the liver anymore.

You can go look at it.

Kalyan Sivasilam: You can look at the other stuff. I think another good example is on a chest x ray, right? Michael, there's this thing on a chest x ray called consolidation. And it's basically when your alveoli are filled with something. The thing is, it could be multiple things. It could be fluid. It could be blood.

It could be pus. It could be cells. And each of them are a completely different diagnosis, even though it looks like exactly the same thing. And so, for example, an AI that is able to. detect a consolidation, get it validated by a radiologist, then goes into your, your hospital information system or your EMR, and then gets the patient history and says that, you know what, I've seen consolidation and this is cough and fever.

And so I'm going to call a pneumonia and get a radiologist to approve that versus I'm seeing consolidation when the patient says weight loss and vomiting, this needs to go into the onco department, right? And then radiologists can validate that. Now you look at the. You look at the quality of diagnosis because AI is able to reach in, understand that there is a difference in differential diagnoses based on past history, and then take action.

And this is where, this is how I think we're all going to get more accurate healthcare.

Michael Marks: Well, I think that's what's really going to drive the value of AI and people will be more focused on accuracy and less focused on, is it eliminating a job or something like that. All right. So let's change subjects entirely.

So a lot of the, the big tech players, NVIDIA, Microsoft, Alphabet, you know, many others, you know, big AI healthcare space in a variety of ways. So how do you. What do you guys think about the build versus buy or partner when it comes to your AI tech stack? Are these companies you're working with? Are these companies that are going to be competing with him?

So why don't you start first, Kalyan? Sure.

Kalyan Sivasilam: You know, in this context, Elon recently tweeted that, you know, you can send up medical images to Grok and it'll give you an answer. And towards the end of this podcast, I'll definitely try to answer why Grok works basically, but why it will not work in the market.

But coming to your question, Michael, I think that if you're somebody who's built AI, you'll understand that. Foundational models are a very rare and very specific thing. I'll be very honest, even on 5C, we, if you're building a CNN for a detection model, I'm either using a ResNet 50 or a DenseNet 121. These are existing foundational image computer vision models, on which we then build our own layers.

Then we quantize them, make sure that they're still meeting performance, and then deploy.

Michael Marks: And are those tools, just sorry to interrupt you, are those tools, coming from these big companies?

Kalyan Sivasilam: These are actually completely open source paradigms, like ResNet and DenseNet. But on top of that, there are foundational models that are now coming in when you're talking about generative AI, right, that is, like we use a lot of Lama, which is again fine tuned on a lot of our own data.

We use Mistral, I think they're doing some really good work. For specific, basic use cases, even a chat GPT or open API would work pretty well. I think that to get to a basic level of accuracy. Using, working with big tech can help, but is big tech going to go all the way and actually solve the problem?

Probably not. And I want to give an example. I mean, Google speaks a lot about the work they're doing in diabetic retinopathy, right? And they're incidentally working in my backyard in Madurai in India. And they keep talking about this model that they built using diabetic retinopathy. I think Sundar spoke about it maybe a year back at Google IO.

You know, it's, it's just one project in this behemoth called Google, and it's a great showpiece. I think to actually solve it, you need to go way deeper, get way more data. And so at the moment, we have found that partnering with a lot of these big tech companies, as you know, I mean, Jensen spoke about 5C and how we're using MonAI to build some of our detection models.

And so I think partnering with these guys is genuinely advantageous. And at the moment, it seems to me that Their healthcare focus is actually towards companies like 5C, like WR, rather than saying, you know what, I'm going to solve this for the patient. I mean, if you ask me, that's the right thing for them to do and the right thing for us to do as well.

Michael Marks: Well, I just wanted to point out to our listeners that, uh, the reference to Jensen was the reference to Jensen Wang who runs NVIDIA. Many of you know that name, but some of you may not.

Kalyan Sivasilam: I mean, he's, he's now a first name basis. Oh yeah. He is. He is.

Michael Marks: Alex, how about you?

Dr. Alex Sardiña: No, I think, I think you're right, and I'm going back to your, uh, Mount Everest analogy.

I think the hyperscalers that, that you've mentioned probably get you to base camp.

Kalyan Sivasilam: Yeah, easy.

Dr. Alex Sardiña: But what takes you to the summit is the passion, the expertise, and the mission to solve that one particular problem. For sure. And I, I think a lot of these big players, as of right now, don't have that focus and that expertise and that passion or mission to get us to the summit.

That's where they, the partnering with companies like 5C and White Rabbit comes in. But you kind of need that basic technology to get you to Basecamp.

Kalyan Sivasilam: You just need the extra dedication to get, and I'll give an example, Michael, like you can take, you can open chat GPT or you can open grok right now and you can download a chest x ray or any basic x ray from radio, PDR, wherever, and it'll predict it accurately.

And now you're going to think, hey, this works well. But don't forget that these massive companies have basically crawled all of the Internet and they have all the openly available data of medical imaging. The moment you show it something that is. you know, from a real life clinical scenario, it is going to fail.

And we've done enough tests around this. And so if you take an image from Google, you know, your grok or your chat GPT has already seen it. And we all know that testing on training data is, uh, is like dunking on a five foot rim, right? You don't do that. So yeah, you want to, you want to, you want to try to play with the big leagues, right?

Michael Marks: Correct. Yep. The way I think about this is that these big companies are really providing the tools and other companies like yours are using those tools at times to help refine the process. All right, changing subjects to our last sort of major subject here, one that I find interesting. We'll talk about geopolitics.

Okay. So we are, uh, We are fortunate to have two very different countries represented here. One is the, the very robust, highly regulated U. S. market, and then the new

Kalyan Sivasilam: upstart, upstart

Michael Marks: up and coming, not very regulated market. So let's talk about how this works for you guys. So, uh, Dr. Estes, let me start with you talk about the regulatory environment, where you are in terms of dealing with it.

And then we'll go to the side that doesn't have to do all that.

Dr. Alex Sardiña: So we, we've actually at White Rabbit, we've embraced. the regulatory environment. And the reason I say that is, you know, most of our competitors have pursued kind of a lower regulatory threshold through the FDA, a process known as the 510k.

Equivalence. Equivalent. Equivalency. Equivalency, right? So you just have to prove equivalency to a predicate product. And that allows you to get to the market, and that's fine. I mean, those products are safe and, you know, in use. But how do you create differentiation? So that's where we embraced the regulatory environment, and we've decided to go down the premarket approval, or the PMA.

A lot of people don't do it. Why? It's expensive. It takes a lot of time. It takes a lot of effort. Uh, it's actually forced White Rabbit to internalize the regulatory position by hiring a VP of regulatory affairs because you have to of

Kalyan Sivasilam: proof is also way higher. The

Dr. Alex Sardiña: burden of proof is way higher. But what is the one thing that people always talk about AI?

Is it safe? Yep. And it's efficacy. So we've decided to choose a path that has the highest regulatory oversight, the highest That will give us the confidence to say we have the safest and the most efficacious product on the market as determined by the FDA and that creates a differentiation from the other players because we can go to market and have something that no one else does and we'd like to talk about this regulatory mode and we embrace that regulatory mode because anyone who wants to follow in our footsteps is going to have to do the same things we did and that gives us an automatic eighteen to twenty month, twenty four month head start.

So. If you're faced with this regulatory environment, embrace it. It's actually, you know, what we've done and I think it's going to position the company very well when the commercialization of the product actually happens.

Michael Marks: I think we'll talk about the commercialization just a little bit. But, uh, Kalyan, why don't you, uh, address that as well?

‍Kalyan Sivasilam: I mean, in India, so recently there is a body in India called the CDSCO, the Central Drug Standards and Control Organization. To make it short, we call it CDISCO. And so the CDISCO in India is the FDA equivalent. And now, a couple of years back, the CDISCO was also charged by the Indian Parliament to oversee AI softwares as a medical device.

And that regulation came out of the blue for everybody. And to some extent, I would say that We've also embraced it very deeply. We understood it's not a regulatory mode like the FDA maybe, but there is a reasonable amount of proof that we even need to show the CDISCO in order to get an AI software as a medical device.

The 510k route doesn't exist because there are no other devices in the sense. But today 5C has the largest library of CDISCO approved AI products that allow us to actually take them to the market. The interesting thing, even with this regulatory environment, right, in India, where CDISCO has come in, there is an understanding within each of the large hospitals that we have deployed in, saying that, listen, CDISCO is a bit like ISO.

It's a checkmark. You need to have it for hygiene. But each of the guys we have worked with, some of the largest hospital systems in India and therefore the world, they all want to do their own pilots. They're all saying, hey, you know what? You have gone and done it. Certified your product with some government agency, and I'm happy, and I accept that as a certification as a feather on your cap, but here are 10, 000 cases, and I expect you to, you know, build me a dashboard that tells me how many of these are normal, how many of these are abnormal, and what pathologies I'm seeing, and let's see, you know, how well this is doing, you know, opposed to this ground truth. And so in India, yes, I think that the regulatory framework is, of course, not as tight as the FDA. I think the two points that I will end with is that one, it is getting tighter. There is now some regulatory approval, and it is very favorable towards Indian companies.

Like if you're an Indian startup like 5C, you get priority access so that you can take your products to market. And that's something, of course, that we appreciate. And the second thing is that, in India, customers understand that products like what 5C has has massive potential. And they basically run their own trials at scale.

Michael Marks: Interesting.

Kalyan Sivasilam: So, in the end, you gotta, you gotta prove yourself, not just at the regulatory level, you gotta prove efficacy. In that center, where you're going to be deployed, and then hope to see some value.

Michael Marks: Well, I think we can all agree that regulation is both good and bad. Sometimes, you know, it's cumbersome and it takes a lot of effort and cost.

But there's value to it because then people feel comfortable with it and all that. In that area, let me talk to you, Alex, about insurance companies, which is a big part of how the U. S. healthcare system works. How are they involved here? How are you working with them? What does the future of that look like?

Dr. Alex Sardiña: That's a great question. For the last Five years or so. I've always been somewhat perplexed by the misalignment of the products as opposed to the end users, right? There's a lot of products and the products have been validated by the FDA and by their commercialization to Improve patient outcomes and make radiologist lives a little bit better, but the misalignment lies in Who's paying for that?

Right. Right. So the misalignment lies, the hospitals happen to pay for it, but yet they perceive the benefit as being with the radiologist and the patients. So that I think is limited adoption. Where I think payers come into play is that they have to start recognizing the value and the cost savings associated with using AI.

And I think that's one of the things that we're doing at White Rabbit is educating the payers that a product that reduces false positives and actually decreases the cost. Of the screening mammography pathway is something that they should get behind. And if they get behind, and if they reimburse, then I think that's going to change that entire paradigm.

So I think they are actually the key stakeholder in driving adoption for some of the products that we're working on right now.

Michael Marks: And what's the state of play now with those companies?

Dr. Alex Sardiña: I think they're starting to realize that. You know, unfortunately, the prior focus of early detection has driven some of that, but they're starting to realize that what's more impactful for them given the life cycle of the patient within their enrollment is reducing false positives.

And I think there's a lot of legislation that's actually in Congress right now, where there's been discussions about expanding the coverage by these payers specifically as it, as it pertains to screening and diagnostic mammography. So anything that you can do to kind of curtail that expense, I think payers are listening.

So I think the tide is changing and I wouldn't be surprised And hopefully we'll be one of the ones to drive this home. That's what I wouldn't be knocking on wood. Wouldn't be surprised that payers actually start reimbursing. There's precedents for this. When the first CAD products came out 20 years ago, they were reimbursed.

Kalyan Sivasilam: I think they came on something called NTAP, right? New technology add on payment. Yeah, correct.

Dr. Alex Sardiña: And then after about 15 years, they started realizing, well, maybe it's not as good. And they started bundling payment, which is what they normally do. Yep. We need to go back to that mindset of if we're going to drive adoption for these much better products.

We've got to reimburse and I think the pairs will be the ones that tip the hat.

Michael Marks: And do you have the same dynamics in India? Are there insurance companies that pay for the who pays for the for the use of these products in India?

Kalyan Sivasilam: I think last count India probably had something like 35 percent covered by some level of insurance.

But you can assume that a significant, I would say 80 percent plus pay out in diagnostics is out of pocket. And so most diagnostics is just out of pocket. And in India, the way this works is all the health providers, health systems that five C is connected to. They are my peers. Right. So they buy the software because they want their radiologists to be more efficient.

They want a faster turnaround times. They want less errors. And so the pitch in India, at least this straightforward. And I think that's a, that's also a very clean way of doing it. Like I am serving the provider in making their radiology departments much more efficient in always in terms of tat, in terms of quality.

And so they pay me. And their deal with the patient is their own, right? And the scan may be reimbursed, but, you know, by insurance, but that does not affect me. For example, even the government is a customer of 5C. We do a significant amount of AI based work for the government of Karnataka, which is where Bangalore is, and from other governments as well.

And the government is a direct payer, like they would cut me a check at the end of the month saying you read these many cases, or this is the license we're paying you for so many EI reads.

Michael Marks: So my understanding is that this is also the case in other developing countries, and perhaps you're thinking about going to other developing countries because they work in the same way, is that right?

Kalyan Sivasilam: Uh, yes. I mean, one is, of course, that the payer provider market is similar, but also that the problems that they face are very similar, right? It's not like the market dynamics. In many of these countries is that there are not enough radiologists. They're all burnt out and they're looking for solutions to make departments more efficient and make turnaround times faster and make accuracy higher.

And so they're all looking for the solution. So as a product, It is definitely something that we can take, not just all across India, which is many countries in one, but all across the world. Yeah, absolutely.

Michael Marks: Well, it's like one of the, I mean, we, we as an investor are invested in both of your companies and excited about that.

One of the things that I like about the developing markets like India is that there's a lot of potential. Pressure to deliver value at a very low cost. Oh, and that's the value that tell me about it. Michael. Well, that's a value that comes over to the more developed countries as we go. So, all right, is a final subject.

And this is fascinating. Um, let's start with you, Dr. S. What, what are the So, going back to sort of the beginning of this discussion about, you know, what people think about AI and, and is it real and is it good or is it bad and all that stuff we already talked about. What, what are the hurdles to overcome, what are the regulatory barriers, are there other things that you need to do and, and perhaps you as, as well, Kelly, in, in order to get a higher level of adoption and acceptance?

Dr. Alex Sardiña: So we've talked about reimbursement already. I think that's, that's. Number one, one of the things that I've always, even before the advent of AI, healthcare industry, at least in the U. S., always lacks commercial industries by seven or ten years in adopting new technology. And it makes sense, right? You're dealing with patient health information, you're dealing with illnesses, higher stakes.

But I think there's been such an overwhelming prevalence of AI outside of healthcare that I think patients are starting to accept the fact that But having some AI in their healthcare journey is acceptable. So I think we're starting to turn the corner. A lot of that also comes with education, educating patients as to the value of AI in their journey makes a lot of sense.

But I think reimbursement is still the one key driver to actually getting this technology to everyone. The world that I would like to see, and Michael, we were having this conversation over dinner last night is providing access to the best quality care to all patients at no cost. Right. So if we have a reimbursement model in place that you can provide this to everyone no matter where they live, no matter their socioeconomic status, that's what we need to do.

Because these, these models are there for a reason and we should decrease any barriers that would prevent all people from having access to our products.

Michael Marks: Well, it's interesting, and that's the first time I've actually heard the, the idea that just the general talk about AI everywhere can be a value to patients who are no longer afraid of it or becoming less afraid of it because they know it's everywhere, part of life.

That's pretty cool. So as a final question for you, Kalyan, how do you think about competition in the world of AI? Because we have these discussions in our partners meetings and all of that, like, how can we So how protectable is this when AI is everywhere and, and, you know, you were saying earlier, at least at simpler levels, you know, anybody can do it.

So how do you protect your business or are you worried about that?

Kalyan Sivasilam: Obviously, I think Dr. S would agree that's something that we constantly think about. I'm going to ask Dr. S. Something that we constantly think about. Today in the world of medical, imaging AI at least, there are a few specific modes.

Doc has already spoken about regulation as a mode. I've also, also touched upon, massive data sets, massive representative data sets. And so, for example, in our case, we probably have like, think two petabytes, two and a half billion images of data across manufacturers, across modalities, across demographics, across diseases.

And so that is one, of course, very valuable. Can it be recreated with a lot of money? It probably can. I think the second thing at least the way I like to operate 5C is, uh, aggression in terms of proving value. And so we are very, very aggressive in getting into large hospitals. Uh, not today in India, but not just in India soon.

And actually proving to them that this is a product that can deliver value to them. And people think about AI as an API in many cases. And I think that is the worst way to think about delivering value through AI. Is that Each hospital, each health system has a different way of understanding how value from AI comes through.

For example, at one very large hospital in India, they want me to take all their scans, triage them as normal or abnormal, and then have another AI, an LLM model, look at the abnormal cases and then triage them into either emergency or surgical candidates. And that is how value comes to them. And so we have a great engineering team that is able to orchestrate all these things together.

And so there's this great article by Ben Thompson of Stratechery about how AI is going to be sold. And I totally agree with him, which is that you can't sell AI as an API. AI sales, at least to enterprise. And we are selling essentially to enterprises, is going to be a lot like mainframe sales, where you need to really get to know your customer, understand their problem, and build a solution for them that solves their problem.

And if you're able to do that, you're building a pretty good business.

Michael Marks: And so, well, of course, everybody has to do a good job with their customers. So we're gonna finish this up with you. Dr. S do you worry about, for example. the hardware suppliers getting into the A. I. Business or other big tech companies you talked about earlier getting into the business.

Is it possible, for example, that A. I. Will get better and will need smaller data sets in order to be good.

Dr. Alex Sardiña: So those kinds of things, it's always something you think about, right? I mean, I think when you're, when you're actually creating technology that can be applied in many different cases, you always have to figure out, well, who's Who's in that space today, who can enter that space because they're larger, they have more data, they have more money.

So, do I see a world where a product like what Kalian is building or what we're doing at White Rabbit becomes integrated as part of a hardware device or a gantry? Absolutely. Yep.

Dr. Alex Sardiña: Um, but I think you hit the nail on the head, which is, you have a question about commoditizing AI. I think AI is so Evolutionary.

It was revolutionary, but now it's so evolutionary. To your point, you have to stay ahead of the game by understanding your customer base and just staying one step ahead and outperforming what the market is doing. Yes, your leap behind products could be acquired and put into a gantry, but then it'll stay there for quite some time.

You just got to make sure that you're moving on to the next best thing.

Michael Marks: So in other words, like every other technology company in the world, you, you can't rest on your laurels. You have to be working on the next product.  

Kalyan Sivasilam: So keep your friends and your enemies really close.

Michael Marks: Well, look, we're going to wrap it up here.

So for me personally, this is great because I love doing a podcast where I learn a bunch, which I did today. And I hope our listeners did as well. A real shout out to both of these companies who are not only doing a great job from a technology standpoint, but a really wonderful job. working to make the world a better place, and that's what we'd all like.

So for Kalyan Sivasilam, thank you for joining from India. And for Dr. S, thank you for joining from Dallas. So our pleasure that we'll call it a wrap. Thank you both.

Kalyan Sivasilam: Appreciate you guys. Likewise.