How AI Slashes Diagnostic Errors in Healthcare

Speaker A: Foreign. Welcome to Reshaping Workflows with Dell Pro Precision and Nvidia, where innovation meets real world impact in high performance computing.

Speaker B: Welcome back to another episode of Reshaping Workflows with Dell Pro Precision and Nvidia RTX Core Pro GPUs. I'm your host, Logan Mahler. So today, healthcare AI couldn't think of anything better that I want to talk about today. So we have a lot with us from AI Doc, so a lot. Welcome to the show. Give a quick background on kind of who you are, where you're from, and then just a very quick snippet of what AIDOC does and we'll jump right into it.

Speaker A: First of all, thanks for having me. My background is from the AI world, so I was new to healthcare with adoc. I served in Israeli Ministry of Defense, headed an AI division. And since growing up I always had this family story which I will share, but I think most people would have something similar and that is my aunt passed away due to a diagnostic error. And when I was, it, uh, was before I was born. But I think that story influence my father and me. So I knew that once I finished my service, that's, that's the thing me and my co founders wanted to do. Just go and jump into healthcare and really ADOC as, um, you know, add through the story. The mission is to uh, improve diagnostic quality and access to care by building the world's diagnostic AI layer.

Speaker B: I know it was a long time ago. Sorry to hear that. Uh, in your. Right. It's something that affects everyone. I mean, not me personally in my life, but I did have a friend whose mom had a bad diagnosis and ultimately, yeah, ultimately ended up passing away. So you're right, it is something that, that touches. So let's. Before I jump into questions, I always like to ask the tough questions. So the misdiagnosis. Right. At the end of the day. And this is not a bash against doctors, not a bash against healthcare in general. Right. But is. It's kind of like a lawyer, uh, because I did go to law school and you go to that extra training to be an expert in your craft and you are held responsible, you know, legally, liably, um, all of that. If you, you know, have a malpractice. Right. And I know, I know doctors are the same. Right. But how do things that are this blatant or blatantly get missed? Like, you know what I mean? Just generally, is it a, uh, doctors are overworked and are looking at 17 scans a day and they're just Overworked, they don't take the time to invest. I mean what, what generally causes it? Because I feel like it's a pretty big thing that like hey, you know, if I'm multitasking at work, you might email me and you get like 30% of my attention. But that is something that should have your undivided attention. Why do you think things get missed at the level they do?

Speaker A: I love that question. So first of all, let me just to give the, the listeners a bit of magnitude. Diagnostic error is, is a very common phenomena and unfortunately it's one of the biggest crises we have in, in the US healthcare. So the data Hopkins, uh, a group of Hopkins published on it is that about every year we have about 400,000 deaths due to diagnostic error.

Speaker B: Are you serious?

Speaker A: Yes. In the U.S. in the U.S. not globally. In the U.S. a year. To give context to how big it is. I know sometimes people don't have context of numbers. It's like nearly how many Americans died in all of World War II over all the time period, right. In one year. And uh, that's happening. So you were asked a really good question. Physicians by the way, are incredible. The pressure they're under is insane. Just to give you. So let's, let me give you a feeling of how the day to day is, let's say let's the radiology as an example because I'll use it more. If you're a radiologist, you're sitting in a room, you have four screens, you have this work list of patients coming in and you're getting one every minute and you just don't have time. There is such a shortage right now. For every trainee that passes radio, we have three new positions. Basically we're missing two. So out of training. So the problem is they're just under this overwhelming. So imagine you gave the email. So imagine you have again, they have more time. But to review a um, much larger shit is that. Imagine you have to respond to an email once every five seconds and every email is very important because you know, the person in the other side has life and death. I mean that is why mistakes happen because at the end of the day we're all human.

Speaker B: So let's jump into this next question. So with AI document, can you market AI that helps radiologists find things that will, you know, underlying issues that patients have. So that's kind of very different marketing, like a very different product than selling, I don't know, project man. If you were a project manager selling project management software like Monday.com right like what is like the biggest thing that you, I won't say had to unlearn, but what is one of the biggest things that maybe came coming into it, you had a misconception on and then you had to kind of unlearn that to build the product in a good way.

Speaker A: So two, uh, things. First of all, just to clarify, I think we're over indexing on, uh, Mrs. M specifically. I think there are all sorts of delayed diagnosis and errors. For example, one of the huge one is backlogs, which is a lot of the value we have is like you have a thousand pairs in backlog, you can get to them on time. I will tell you where I one of the misconceptions I had and when we started we thought it's going to be an efficiency play. It's all about how AI can make you more productive and all of that stuff. And uh, we hear it all the time in AI that was just not the main value driver. At the end of the day, productivity is very hard to effectively monetize because, okay, let's say you read whatever 20% more like. So, so what you're gonna do 20% more, you're gonna take a coffee like it's unclear what actually is. And we found a way to do it is talk about quality. And that was a big surprise because uh, everybody tells us nobody cares about quality. I care, yes, as a patient you do. But I think there is this uh, truism that again I don't think is true. But when you go into healthcare, people are very cynical about quality and over index on ROI and thinking about this because again they become desensitized. Think about a health system that always have to make life and death decisions with, with money. But I actually do think that quality matters. And there are exact links. By showing them how you improve quality, you can actually improve the system downstream. So for example, if you catch the patient earlier, you may treat it earlier and have a reduced length of stay, which is a big operational metric. So the fact we actually shifted from productivity, which everybody told us is going to be a productivity play, to quality one was a really big.

Speaker B: That's really interesting. Okay, I love that. So that kind of ties in my next question. So AI docs, you have kind of your first read AI, right? Like drafting radiology reports. So from my understanding it just got kind of an FDA breakthrough device designation. First off, what, what is that? And then what does that mean? Is AI flagging? Like, let's say I'm um, using example, but like is that Hey, I have cancer writer report saying, hey, this is what kind of cancer? Like what exactly? One is what is the destination and two, what is the first read? AI, what does it do? If I'm a patient, what does it do for me?

Speaker A: Look, we've been at this for almost a decade at this point, and for me it feels like a new world has started with these new devices and foundation models. So a moment of context when you do. The problem with clinical AI is you have to be extremely accurate to support physicians, because physicians are really good. So the level they expect is also very high. And otherwise it's junk. So in the past you could do it, but you had to do it for one disease at a time. So literally every company on the face of the planet, they would build these clinical AI products that focus on one disease. It could be cancer, it could be stroke, could be bleeds, whatever it is. And that was obviously very limited in scope. ADOC is a company with the most FDA clearances in the world for this category.

Speaker B: That was my next question. It was 18, right?

Speaker A: It was 18. That's a really good. Yes, it was 18 in the beginning of the year.

Speaker B: Okay, I had 18. That's my next question. My next question says, literally you have 18 FDA cleared algorithms.

Speaker A: I had 18. And this comes to this point, I had 18 until the foundation model came. So we've decided to make the decision together with Nvidia and um, aws, to build our own diagnostic foundation model, which is, think about it like a chatgpt that can look at an image, or maybe Claude would be better. Claude. It looks at an image and finds every disease all at once. So that thing is extremely powerful. That model is called care. And on care we build first read. So first read, think about it as the cowork, like the cloud cowork, that is building on the opus model. Right? So it basically takes the image and. And then finds every finding in it and then drafts a report. Obviously still a human to review. It does not replace physician judgment, but it actually does clinical work for the first time, by the way, in medicine. To the best of my understanding, this is possible that we have an AI that actually does diagnostic work, do the breakthrough device. So breakthrough device, that's a really powerful regulatory mechanism. So FDA recognizes some technologies are immensely impactful on human lives, but also are very innovative and require some sort of thinking and paradigm. And we cannot engage that regular without breakthrough device. You could engage fda, but it's a very slow process. Every iteration takes like six to nine months. This breakthrough device Basically gives you a program by which you can very rapidly iterate with the fda. So it's not, to be clear, FDA is not picking favorites. It's a program that everybody can access, uh, if they meet the criteria. But we have this ability to engage with the DA and work very closely to and ideally to bring this device to market much more rapidly. And I think it shortened years out of the time to market.

Speaker B: That's very cool. So kind of next question. I mean you mentioned Nvidia, right? So I know inventures invested in you and I a while back or fairly recently. So to build your own foundational model. So kind of a first question is what has the Nvidia. Well, I mean other than providing you, you know, your ABC or whatever round of funding it was, other than the shot in the arm with some money, like what has Nvidia, uh, really meant to AI doc, right? Like uh, are they providing resources, are they know how, like opening doors for you, what have they provided?

Speaker A: Bottom line is yes to exactly these three if I would have to match them. So it's know how and support. So for example, um, maybe again a note of context I will tell you. So when we made the decision to build our own foundation model, it was a few years back and you know, November 2022, ChatGPT came to the scenes. We all recognized something is different we have to build. But back then we were an earlier stage startup. We knew it's going to cost a lot of money. And I will tell you when I came to the board and tell them, hey guys, I'm going to be spending hundreds of millions of dollars that I don't currently have on something that I don't know if it will work, but I think we need to bet the company on uh, was a very scary uh, decision if I'm being honest. We knew if this doesn't work we could basically fail. Uh, but I also recognize that to really get to our mission of improving diagnostic quality in a broader sense, we have to go faster. So that's made the decision. Nvidia was a really key partner because of their ability to help us optimize the resources. The cost of for example doing a lot of dependent optimization on the training. These are really, really, really large data sets. So any ever training run could cost, you know, if I'm managing the millions. So thinking through the architecture, thinking through the training, they've been immensely helpful. They're also a great partner in opening doors I would say. But obviously the big thing that they're bringing uniquely is Their ability to really help you kind of optimize the technology, the training. Um, and they've been awesome at that.

Speaker B: Okay, so let's go to this training of the foundational model. Right? Is I. And you're going to laugh at this, but I had a coworker. There's someone. I lead AI for Dell. Right. But there was a coworker of mine that does health care, life sciences. She was at what's it? Hims, I think hims. And she was like. And I'm fairly technical. Um, more. More than this person. And she was like, hey, I want to be able to show some radiological scans and that. It, uh, just kind of show people the general idea. Obviously, this is before we met y', all, obviously, next time you're being pulled in, clearly. Um, but I built a little thing, right? A little web app that did it. And who knows if it was right or wrong, but the idea was there. I guess the question is, when building a foundational model, like, are you licensing it out to other people? Like, is it is Claude? And I got. I hate to even admit this is a local AI show. Right. But to Claude is. I'll just say Claude is Claude. Right? Opus man is just great. And it'll do everything by training. Like, what is the strategic advantage other than you own the ip, you own the model. Is it actually better on benchmarking stats? Like, what. What is the real value?

Speaker A: Yeah, you're asking a really good question. And I will tell you, if I didn't have to train it, I would definitely not train it. So it's not. There is no, like, big business benefit from, you know, incinerating hundreds of millions of dollars. But I will say the thing is that in healthcare and clinical AI specifically, is that accuracy is immensely important. And this is a slightly different. Not slightly. It's a different task than the typical LLMs are trained on. Right. You think about it as, um, so if you have a CT scan, um, with all the clinical information, but let's look at the image the CT scan contains. Could be a thousand images and you have, like, 10 pixels in the wrong color. And that is a disease. It's like, um, imagine you need to do Where's Waldo? You need to train an AI to find Waldo, but there's only one waldo in, like, 10 books. So that is roughly what we need to do. And it's just a different task. And the accuracy level we need to build is. So the data is different, the task is different, and the accuracy requirements are completely separate. So we've tested everything on the face of the planet. I don't want to mention the models just for their sake, but they were not even close. It was 100x the error rate, not even 10x, 100x the error rate we thought is reasonable. So we had to work on really optimizing and building the model in the way that fits the market. And that's the only way. And look hard to know. And I don't want, um, to with humility, but I think it's a very different space. I don't know if automatically the frontier models would like one day just start doing this. It's such a different task.

Speaker B: Well, it's a huge liability as well too, right? Like, I mean, there's a difference. And this kind of comes to another question. There's kind of a huge. Well, I mean, there's a huge liability with part one. Part two is it's probably pretty easy for the folks at Anthropic to say, hey, here's a bunch of CT scans. If you see any, pick if, if it truly comes down to, and I've never had a scan, hopefully never get cancer, I'll use my mom as an example. She did have breast cancer. She did survive. But if it really comes down to, hey, a couple of pixels or a different color, it'd be pretty easy for them to say, hey, here's a bunch of CT scans. Look for pixels that are different colors, but that could be a speck of dust. That could be all types of crazy stuff, right? Like, this is much different. So I, my kind of question is, I'm assuming when you're training this, you know, foundational model that you all have, or is that you probably had to employ quite a few doctors and people that were very good to be able to say, hey, this is, and I'm not a cancer expert, this is breast cancer, this is carcinoma, this is, you know, all these types of cancers to really have a true, accurate data set of what this look like. And I'm assuming that was a huge part of the process.

Speaker A: Huge part of the process. And also I will say the, the data for the training is one thing, but the validation is even crazier because you need to know how accurate it is. So I'll give an example. We got NFDA clearance on the foundation model for the CT of the abdomen that finds, uh, 15 different diseases in the abdomen area. The specificity, uh, of this was 99.7%, meaning the false positive rate was 0.3. Think about what it takes to even measure something is 0.3% accurate. I mean, the data sets we had to build to do this, um, is insane. So you're right. I think the validation and accuracy, again, it's a whole core competency. It's clinicians and what matters. You also need to get into the definitions of every disease because it's not even like a binary thing. Like, it's not a. Yes, no. Cancer is a type of cancer in a specific setting. It's benign and malignant. Like, there's all sort of these nuances that sound, you know, in theory, it's yes, no brain bleed.

Speaker B: In practice, that's a data set for all that. So you've got, hey, here's the disease. Here is cancerous, here is benign, here is stage 1, 2, 3, 4. Like, oh my God, that's insane. That's insane. So, so Bridge framework, uh, which you released open source with Nvidia, it's basically correct, um, me if I'm wrong, but it's basically you and Nvidia telling the market, hey, here's how the industry should evaluate clinical AI, including our competitors and us. So it's, it's kind of like, you know, if I put my daughter in charge of, you know, cleaning the house and then I let her self check on how good of a job she does. Right? Like, you've kind of set your own benchmark. And I'm joking, but I am kind of serious. Like, have you one gotten any pushback or fight about it, uh, from other people saying, hey, you're using your own thing to grade yourself. But, like, what is the real value that the Bridge Framework brings to the market?

Speaker A: I had a realization. Who's the people who's doing the most benchmarking on Opus or the model? Who's building the model cards and doing the alignment checking? It's entropic. They're releasing these hundreds of pages, model cards. They're the people that understand the beast, the created and how to test it. So to some extent that is natural. And I would love for somebody else to take from me and say, okay, we will track it. But it's very hard to build that core competency of understanding that. So we believe we have to do it to some extent. But Braid Framework is slightly more than that, just to clarify, because the idea is to really foster the ecosystem. So the problem is that there is one aspect which is the model, but there is the other aspect which is the integration. So ADOC is kind of a company, we both have the models, but we also build the equivalent of A harness, a healthcare harness, which, uh, we call aios. It's basically the platform that contains all the monitoring, the governance, the workflow integrations. We basically. And now there are a lot of companies that think about, like, Netflix, right? So we have the Netflix streaming service and you have the Netflix originals. So we also build a lot of Netflix originals, but we also have the streaming service and a lot of shows or companies want to come on board on adoc because we're the most well distributed clinical AI company. So we basically said, look, if you want to come and do this, either you're a platform company or like a Netflix equivalent or a solution, you have to adhere to specific standards for that integration and workflow. Because it may sound simple, but there is so many nuances and details at the end of the day. Um, so that is really on top of the testing. That is what it allows to build, to say to people, hey, these are the specs of how you go through putting something into an ecosystem, if that

Speaker B: makes sense, makes total sense. And that's kind of my next question. So aios. So it is kind of like the Netflix of, you know, clinical healthcare AI, which is kind of different than just AI Doc in general. Is anyone ever confused what your core competency is? Because being the Netflix of clinical healthcare AI is very different than, hey, I'm building my own foundational model and putting my model here. You know what I'm saying?

Speaker A: I know what you're saying, and you're right, it is confusing. And look, part of the difficulty is that in healthcare and in health systems, a lot of these components are missing. It may sound like, I wanna do all of these. I don't. I'm perfectly happy to be either or if the other side existed reliably. I think we just realized we had to build both. So think about the early days of Netflix, but there were no TV shows. So we were like, okay, so we have to build the TV shows, but we also have to build Netflix because otherwise people can't see. And it was all based on, by the way, market feedback, because we started by building the show, we started by building our own AI solutions on specific diseases. But we saw that they didn't scale, so we had to build the infrastructure and then open it to others. So it is confusing to some extent. But at the end of the day, you know, we sell, I think, the Netflix analogy service. Well, Netflix is both a very good content house as well as a streaming service. And yes, those are different core competencies. We had to become good at both. But um, at the end of the day we think they're synergetic, um, and adding value from one another. Because I know to develop AI, I know how to monitor it and how to integrate it, uh, and vice versa. I know how to build my algorithms to make sure they adhere to that. So it does help.

Speaker B: It makes sense. The only thing I would say about your, your analogy, which is different, you built the show model kind of first, then you built the Netflix platform. Netflix was opposite, right? They had the platform, then started doing it. But it's all the same. So a couple of use cases, right. And these hopefully aren't ones that uh, I mean I found them on the Internet, so hopefully they're not proprietary. But for example, you work with Sutter Health and Advocate Health, right? Both enterprise kind of, you know, enterprise wide deals. Tell me a little bit like if we have someone who is a hospital administrator, part of a large health group or a hospital, what does that kind of look like? I'm not talking about pricing, but is it. Obviously I would assume there's a pilot phase and then there's a scale phase and then there's a full deploy phase. Tell them what the, the general kind of process looks like if uh, if someone was interested in purchasing aidoc.

Speaker A: So the first thing, even before you get started on anything is to scope out what are you trying to achieve. I think that's a very important step. And so the first thing is we model the outcome. So we look at the data together, we say look, this is what you would expect. This is the clinical impact, this is the financial impact, this is what we're trying to achieve. Surprisingly there are less pilots at these phase. But you would say let's scope out a package of diseases. At this point in time, the typical deployment will be 20 different disease states. Um, and it's important to still pick diseases because there are a lot of hospital still works on a disease based, you know, there's a care pathway. If you have a brain bleed, that's what happens. So we, we have these whatever we, we scope it out with them and then we, and then we deploy basically. The funny thing is that the more we talk with people, they recommend go full enterprise the fastest you can because otherwise you have variability of care. So imagine you have one hospital that you get treated this way and that way we're totally fine. Also doing a pilot in a region, um, and then scaling. But I think generally like Sutter, I think as an example that was by the way insane. Um, we went from first meeting to full Validation, safety and enterprise go live within 12 months. Like from the first meeting we ever met to that, which is insane phase for health systems. Um, and again enterprise wide. So I think tens of hospitals, thousands of clinicians, millions of patients. Right. That's very rapid. So health systems do that. Ah, Advocate, for example. They did a region first and then scale enterprise wide. But Advocate is I think, the third or fourth largest health system in the nation. So they're humongous. So yes, hopefully that helps. So you go in, you scope out, you create a model, and then you basically deploy and measure and then iterate.

Speaker B: I love it. So Wellspan CEO said that you helped, um, their radiologist scan over 200,000 cases in one year, basically cutting, you know, critical diagnostic delays. Like, I mean, that's a great stat. But like, give me kind of. We were talking about ROI a little bit. And you don't have to give me the exact ROI because that's like patented, that's trade secret. But like, what does that mean? That's a number. Like, how many people did you potentially save? How much faster did you get? Do you have any of the stats behind that?

Speaker A: Yes. So, by the way, well spend is a great story because it massively expanded since. So I think we're at a point of 1.6 million patient studies, uh, at this point a year. Yeah, it's pretty big. Pretty big. And look, at the end of the day, I would have to look at Wellspend specifically, but generally you would look at all of the different value levers. So we talked about backlogs, for example. So what typically happens? And I don't know, Wellspan, Amber specifically. But let me give you an example from another health system. We find there are certain diseases that even if you're an outpatient and you're not expecting it, you need to get your care within typically 24 hours. An example is pulmonary embolism. So let's say you're.

Speaker B: They'll kill you quick. That'll kill you quick. If it blows up, you're done. Done.

Speaker A: The problem is today there's such a physician shortage. So even since you did the imaging, you could wait days, if not a week to get your. To get your results. Not always on every. Not everywhere, but let's say 10% of the patients. And we can look at the data and say 10% of the patients would take typically over 24 hours. With this technology, AI flags them to be read within minutes. So while the patient, like, imagine the different experience. You're getting a call a day after potentially Getting a call. If you're, if you're able to answer the phone, okay, a day after you have a critical disease or you're in the facility, they're finding something while you're in there and taking you by the hand to the ad.

Speaker B: That's. I mean, it's a completely different experience, right? Especially something like, uh, pulmonary embolism. Because I'm not a doctor, not a healthcare expert, but it's one of those things. If it goes, it's over. Like, it's just very, very quick. So, I mean, I think that's great. Like, I think you're right. It is not necessarily about the volume. It's like how rapid it is. Like, that's the key. Like, I. I mean, honestly, I'm just kind of sitting in shock because, like, I didn't even think about that use case. It's true. Like, I mean, you literally see in the movies, right, where you know, they go and get a scan and then they rush them to surgery. Well, that never happens. Like, you know, I, like, I literally back it. I mean, this has been, uh, many years ago, but before AI but like, even with the scan of my appendix, it was like, oh, yeah, we might have. Let's go home. And then they're like, oh, my God, please get him back up here. His appendix is about to burst. Do you know what I mean? Like, it would be such better going in and just being like, okay, Logan, we're off.

Speaker A: Absolutely, absolutely. And by the way, the beautiful thing is that, like, this is. This is what physicians want to do, right? They want to give the, like, they've been trying to get to give the patients, uh, the care. They're just so overwhelmed. So these tools really help them do that.

Speaker B: I love that. So you said by 2030, every complex diagnostic decision should be supported by AI. So between now and then, which, what is that? Four, three, and let's call it three years and four months. What actually has to change for that to happen?

Speaker A: This is why foundation models for me are so transformational, because they actually allow us to cover that. My actual, My real belief, if you ask me. And look, I can't commit to anything. We'll do our best, but my real belief is every. We will cover every disease on at least a CT or X ray in the next 18 months, which is something that is almost would like. It would have been ridiculous to utter the words even two years ago. So what needs to happen for us to cover everything is obviously, you have to build this diagnostic AI model that has to be able to go through modalities, analyze the record the image, uh, the pathology, the genomics. Right. So you have to build the model. Then that's one you need to validate it and get it through regulatory clearance. And the second thing, you need to build the piping that is AIOS or similar platform. So we would have to be able to spread it across as many institutions as we can. Right, so cover the full nation. And then the second thing is build this model that can find every disease. But I think if you do that, Bill Gates said have a PC, uh, on every desk. I think we can actually have an AI on every complex diagnosis.

Speaker B: I love that. So, last section before we start wrapping up. I love doing rapid fires. So it'll be a quick question, Give me your answer and maybe a one sentence. Why? Um, so first off is radiology or cardiology, which one is harder for AI Doc? Like those scans kind of equivalent.

Speaker A: I mean, I really, I'm not trying to say. Uh, you know, I'm. I don't want to pick favorite. It's just like they're, they're both difficult for very different problems.

Speaker B: Okay, that's fine. You have to pick both. FDA clearance or peer reviewed study. Which one matters more to buyers?

Speaker A: FDA clearance. It's necessary to sell. So unless I want to be in jail.

Speaker B: We don't want you in jail. Uh, CIO or chief Medical Officer. Who's the real decision maker?

Speaker A: It's most health systems, the CIO makes more of the budget decisions. In some, the CMO does. So it is heavily depending, but like more common. The CEOs makes the kind of the purchasing decisions.

Speaker B: It makes sense. Like it's, it's no different than me recommending to our CIA or our CTO or cio. Hey, we need to do this, you know, because they're just making the purchases. There's nothing wrong with that. So brand trust or clinical evidence? With a new customer, what opens the door first?

Speaker A: Brand trust.

Speaker B: Okay, I like that. New York or Tel Aviv?

Speaker A: I like Tel Aviv more.

Speaker B: It's okay. I get it. It's fine. It's uh, fine. Is there any city in the US that you like more than New York?

Speaker A: I love in Miami. I love Miami.

Speaker B: Who doesn't love Miami, man? Who doesn't love Miami? All right, well, uh, this was fantastic. So we're going to go ahead and wrap it up, but pretend someone just came into the episode. Now give them kind of what do they need to walk away with and then we'll wrap up the episode.

Speaker A: The key nugget that we're on the precipice of the biggest change in healthcare that has been happening in decades, and that is foundation models. I think we're, uh, you know, if it's five years or 10 years, doesn't really matter. But in a few years, we will have AI on every complex diagnosis. And what it would do, it would do two big things. First of all, it would massively reduce the diagnostic errors we're talking about. Right? So the 400,000. We mentioned, 400,000 deaths. I think we can massively reduce that number. And two, it. It will really massively expand access to care because physicians could do a lot more, so you wouldn't have to wait two weeks for diagnosis. And I think this is coming now.

Speaker B: I love it. This has been such a good episode. So someone's listening. They want to reach out to you or they want to reach out to aidoc. Where can they find you? Where should they go?

Speaker A: My email is perfectly fine. I'm happy aladoc.com.

Speaker B: yeah, perfect. So with that, really appreciate you having, uh, you coming on a lot to the episode. It was just fantastic. And with this, you know, another great example of what AI can do and how AI is changing our lives for the better, uh, whether we kind of know it or not. So with that, until the next episode, this is Logan with reshaping workflows with Delpro Precision, Nvidia RTX. GPUs signing off. Till the next one. Do what you want.

Speaker A: Do what you want.

Speaker B: This podcast was produced in partnership with Amaze Media Labs.