#443 AI in Healthcare: Dr. Robert Toth on Medical Imaging, Robot Surgeons, and Predictive Medicine

[00:00:00]

Mehmet: Hello and welcome back to an episode of the CTO show with Mehmet. Today I'm very pleased joining me, Dr. Robert Toth. Robert, the way I love to do, you know, the intros, as I was telling you, I keep to my guests to tell us more about themselves, you know, [00:01:00] so your journey, you know, your experience and what you're currently up to.

Mehmet: And then we take it from there. So the floor is yours.

Robert: Great. Great to be here. Um, my name is, uh, Robert Toth. Um, I have a doctorate in biomedical engineering with a focus on medical imaging AI. Um, I'm an entrepreneur. I'm a founder. Uh, in the last 11 years, I've been running my company Thetatech. Data tech dot a I where we build custom a I and health care tech systems for our clients.

Robert: We work with a lot of universities. We work with medical device companies, and we are just obsessed with the latest technology and latest a I stuff. So that's a 30 seconds about me.

Mehmet: Okay, great. And thank you again, Robert, for being here with me today. Kind of a casual traditional question. Sorry. Um, why biomedical medical And why, you know, AI?

Mehmet: So, so what, you know, drove your attention to be in this space? [00:02:00]

Robert: So, I've always been into the AI side of things, um, machine learning, I started learning about all this stuff as an undergrad engineering student in 2005 2006. Um, I thought it was the coolest thing in the world that computers could look at an image and find some patterns that detected cancer.

Robert: Um, I wanted to focus any technology skills I had and coding skills on something that could directly impact patients and people and like save some lives. And so biomedical engineering is the perfect use of machine learning to me in a real meaningful endeavor. I really just, I want to see robot surgeons.

Robert: I want to see. Cool, high tech AI saving lives, letting people live longer, and it was just a perfect, perfect focus for me to do biomedical engineering.

Mehmet: Absolutely. Now, you mentioned something about saving lives. You want to see all these innovations, you know, uh, being shaped. So we know one thing [00:03:00] usually, Robert, about, you know, healthcare in general, like it's highly regulated.

Mehmet: Of course, it should be. Um, it needs also high accuracy. There is no space to get wrong. So how you know, you see, you know, with a I going so fast and knowing like the regulatory, you know, challenges off of health care. So how are you seeing these? Going hands in hands, I would say, and where do you see the AI can make, like, the fastest impact, um, today?

Robert: So, the FDA recently released a guidance document on their view of the future of AI in healthcare, obviously a highly regulated industry. Um, a lot of the focus is about testing how generalizable these AI models are. So if you're doing AI to predict something, if you're doing it to make a diagnosis or to recommend a certain cancer drug, because [00:04:00] the AI says this will work for this patient, they want to know, did you apply that AI to people in another state, another country with different genetics, with different dietary habits.

Robert: There's a huge focus on the generalizability of these AI systems, um, and I think that we're finding. A lot of healthcare is going to easily incorporate. The generative LLM based AI to help with the paperwork. I think the lowest hanging fruit is just the onerous paperwork for insurance billing, uh, scheduling, you know, dealing with burnout, transcribing doctor's notes, giving summaries of, uh, patient meetings with doctors.

Robert: Uh, summarizing the patient journey in an LLM generated report. So, I think every health care provider and hospital, et cetera, is going to have AI generated, AI summaries everywhere. So, the [00:05:00] regulatory people are going to have to say, Hey, if the AI report doesn't include some information, The doctor is still liable and the doctor has to double check what the A.

Robert: I. Says. So I think the regulations are one going to focus on how generalizable this A. I. Is. But secondly, who's accountable and who's double checking the A. I. 's work? I think that's going to be one of the most important considerations.

Mehmet: Absolutely. Now, again, talking about A. I., Robert, and something that comes always, regardless which, uh, vertical we're discussing is, How people gonna start to adapt, you know, uh, and adapt at the same time to this new fact that AI is with us.

Mehmet: So when it comes to, you know, the people who work in healthcare, let's say starting from the nurses, starting, you know, from doctors. How are you seeing their excitement or maybe a little bit pushback? So what are like some of the things that you. [00:06:00] You started to, to monitor and, and, uh, notice when it comes to adoption and adoption, I would say.

Robert: Uh, I think the adoption is Tricky when the doctors don't want to be replaced, the nurses don't want to be replaced, everyone's very scared of being replaced. I think it's inevitable that the large companies like Epic, for instance, who makes the EHR reporting software, is just going to start seeping in AI summaries of things and AI generated content.

Robert: Um, I think nurses and doctors aren't using AI tools to double check themselves right now. They're not going to chat GPT. They're not going to perplexity. They're not double checking. Hey, did I get this diagnosis right or other, other considerations? Um, so I think there's a little bit of subtle pushback of, I'm not going to use this.

Robert: I'm going to rely on my knowledge, but I think the adaptation has to be. Hey, you know, read over the A. I. Summary of the zoom [00:07:00] meeting. Read over the A. I. Summary of my transcript with the patient. See if it did a good job. And I think just getting used to those like A. I. Generated diagnosis. A. I. Generated summary.

Robert: A. I. Generated whatever. Um, you know, I, I think the people who are getting used to it and embracing it are going to be the ones who stay on, uh, on the cusp of it. And the ones who are too scared, it's going to replace them. Um, They're kind of digging their own grave in a lot of ways if they're not embracing it.

Mehmet: So this is the same applies to any other field, I would say, right? So always we say like, yeah, you need to, you need to teach yourself how to use the, the, these AI tools, otherwise, you know, you might lose your job to someone who knows how to use these tools. It's not about replacement. And I like this fact, Robert.

Mehmet: Now you mentioned something in the intros about your excitement of having robots in the operation rooms and all this stuff. How far are we from this? Like, Oh, we're pretty

Robert: far. We're pretty far. Dexterity is [00:08:00] not a solved problem yet. It's much easier to have an AI churn through digital EHR records than to have a robot that is able to automatically do surgery or something like that.

Robert: Now, I think things like the DaVinci robot and there's other tele operated robots are going to be a step in between here in full automation, where you have a doctor around the world doing remote tele operations. Then slowly, pieces of that will be automated. Okay, well, you know, make the first incision, automate that.

Robert: Then The doctor goes in and looks around and does something. Okay, you know, stitch them back up at the end. Automate that. Like, they're gonna start to automate little pieces of that gradually as robotics gets better. But, you know, robotics is hard and in surgeries you need a human right now at least to handle unexpected situations.

Robert: Oh, no, there's a, you know, something's bleeding. What do we do? Like, right now [00:09:00] you need the robot. to not take over in that situation. You want a human in the room. Um, but I think it's, I think it's just inevitable.

Mehmet: Right. So we talk in a, you know, in other fields, again, about the concept of co pilot, of course, the, the term Microsoft adopted it in, in, in their AI thing.

Mehmet: Um, so it's kind of like, I see it myself. To your point, like always we need the doctor to be, you know, maybe supervising what the robot is doing so we can think about it maybe as a co pilot more than an autonomous, right?

Robert: Yeah, it's like when the doctor says to the nurse, hand me the scalpel, he can say to the robot, make an incision there, you know, or, you know, reposition yourself five centimeters to the left.

Robert: Um, it's like he's using it as. It's another person or another being in the operating room that he can direct or she can direct and kind of use them as like a copilot is a good example, like, okay, I'm going to, you [00:10:00] release the landing gear, I'm going to, you know, get the plane to drop altitude, whatever it is, having the robot in there or a team of a suite of robots that could do things with high precision repeatedly, like stitch someone up.

Robert: You know, that's something that a robot could easily do if a doctor tells him to or, you know, clamp this area. Does he really or she really need to be telling a nurse to do that versus a robot? Some of those tasks are simple.

Mehmet: Right. Now, I know also for a fact, while preparing for today's episode, like, uh, one of the main things you specialize, you know, in is, um, AI for medical imaging.

Mehmet: Right. So what are like some of the most exciting breakthroughs I would say you have seen recently, you know, when it comes to, Reading maybe x rays or like radiology results. And if you want to compare, you know, the way today's AI models, especially in this field to [00:11:00] maybe some of the primitive traditional approaches we used to use for diagnostics and, you know, kind of reading the results fast and, you know, telling the patient, okay, you might have this, you might have that.

Mehmet: So where are we heading and what are like some of the exciting things you can tell us?

Robert: I think the exciting things are, in radiology at least, a lot of 3D models are coming out. So, a lot of the research on, you know, facial tracking in videos and stuff like that, it's all 2D computer vision. And so, Project Moni by NVIDIA has done a great job of collecting a bunch of really good open source 3D models.

Robert: Uh, Hugging Face hosts a bunch of 3D medical imaging models. Because when you take an image of the human body, uh, you take an MRI scan, that's a 3D native thing. You're taking, you know, X, Y, and Z of the skull area, for instance. And a lot of the exciting new models coming out [00:12:00] have adapted the 2D models like, uh, Facebook's SAM model for segmentation and they've adapted it to medical imaging and they've extended it from 2D to 3D.

Robert: So, I think some of the biggest exciting things are automatic segmentation of organs, 3D renderings, uh, augmented reality where The doctor or surgeon, whatever, whomever is doing the work in the operating room, can see an augmented view of the cancer or the blood work or the radiology imaging. Um, but I think the biggest thing for the radiology and medical imaging in general is just the proliferation of these open source computer vision models and other fields just seeping into medical imaging and people converting them to either.

Robert: You know, 3d or higher resolution 2d in the case of digital pathology. So just all these open source models exploding onto the scene. That's great.

Mehmet: Absolutely. It's, it's fantastic. You know, and you know, we really, really, I tell people something [00:13:00] because I come from a technology background, but really we are living in such exciting times, especially when it comes to things, touching people directly and, you know, their health, of course.

Mehmet: Now regarding what you mentioned about the medical imaging and you know, diagnostics and all this. How do you see that intersects with, you know, kind of, I had just released a podcast episode about, you know, buildings and predictive maintenance. Of course, if we want to take it to the healthcare, like predictive medicine.

Mehmet: So is this something that you think that we can achieve with the current AI models that we have today? So maybe I don't know. You take maybe blood samples from me, you analyze my DNA and couple of other things and then you tell me, me, you know what? Based on what we have seen is you need to take care of 1, 2, 3, because if you don't do so, the pattern, the LLM, whatever tells us like you might have risk on having these diseases.[00:14:00]

Mehmet: So, and maybe there will be some ways to correct that also as well. Am I talking too much futuristic? Is this something which is realistically. possible. Are there some progress in that field, Robert?

Robert: I think so. I think you're not too futuristic. I think the predictive personalized things. So I think there's two categories.

Robert: One is the FDA approved devices that say, Hey, based on your genetic sequencing, you are at risk for certain diseases. Therefore, you know, do these certain behaviors or get screened early. That's one thing. I think screenings are going to proliferate. Um, if everyone can get a cheap CT or MRI screening or heart disease screening and AI says, Hey, you're, you know, this is your percent risk of heart disease or lung cancer or whatever it is.

Robert: Screening is going to go up as people get cheaper AI models. Um, I think there's also a category of health and wellness that are not FDA approved devices. So you're thinking like. [00:15:00] Fitbits, you're thinking you're wearables, heart rate monitors. Uh, pinpricks for a little, you know, blood analysis, uh, the company Throne that analyzes the waste in toilets with computer vision, and I think there's going to be a lot of home, uh, solutions, consumer based solutions, patient based solutions that say, hey, you know, we monitored your, Grocery purchases, and we monitored your blood or we checked your blood work and you had a screening that checked your liver and the level of fat on the liver.

Robert: And because of all these, you know, you should adapt your diet in the next 3 months to, you know, eat more carrots. I don't know whatever it is, but I think this personalized like health and wellness at home consumer owns the data is going to be a huge industry that helps predict, uh, you know, what. Okay.

Robert: People are at risk for and potentially offer suggestions and as long as they don't report scientific findings you know, we'll see if the FDA allows the AI to just let a consumer with their own [00:16:00] Fitbit at home get a You know a health prediction that says hey, you know If you don't hit 10, 000 steps a day, you're going to drop dead of a heart attack in 10 years, you know, give or take whatever way it phrases that.

Robert: I think there's just going to be a huge amount of consumer based, uh, goods that are AI generated, uh, predictions.

Mehmet: Yeah. And one question just popped up in my head because, uh, you know, OpenAI brought the deep research model, um, two weeks ago. And I start to see doctors specifically, you know, which I follow, uh, very closely, uh, on different social media platform.

Mehmet: And they were talking about the accuracy and, you know, the speed that specifically, you know, the OpenAI one, and I'm sure like more models would come and follow, especially when it comes to. You know, for example, reading, uh, a specific case, let's say it's a cancer case and maybe sometimes they are not sure, should we go with a chemotherapy?

Mehmet: Should [00:17:00] we go with radiation? Um, where are we heading in that space, Robert, when it comes also to the research part and you coming from an academic also background as well, you have a PhD in biomedical engineering. So, Can AI also help us on the research part of that, which can translate later into practical aspects?

Robert: I think so. I think the best thing it can do is scour PubMed and read the abstracts of a hundred relevant papers and then read an individual case report or grant that it needs to edit and use all this outside knowledge. I, I just, Think the deep research, you know, deep seeks are one doesn't do deep research, but it does reasoning perplexity.

Robert: All this stuff is going to help researchers make connections that they haven't made. If they ask it the right way, you have to basically say, Hey, you know, deep research summarize. [00:18:00] The latest 100 papers on glioblastoma and then tell me, given that information and given this one case, what do you think is the best course of action?

Robert: Or what do these 100 papers have in common? Is there some biochemical we missed that was an obscure research paper that I didn't have time to read or the professor's students didn't have time to read and deep research would go out Find the needle in the haystack little tidbits of information and give you a report that the researcher didn't know that doesn't mean the researcher is Replaced it means the researcher has another tool to learn a lot more a lot quicker

Mehmet: cool Now here's my other question to you and Again, I'm relying on your academic plus also practical background, Robert.

Mehmet: So, and this is not only in healthcare, but in many situations and because I talk to a lot of people in the healthcare industry. So what they used to tell me is sometimes you do something in theory, it works [00:19:00] perfect. But when you go and do the clinical trials, You know, it's completely different results.

Mehmet: So Are there like some best practices while doing this research to make sure that when we jump onto the clinic It's gonna work the same way as we have found out during the research

Robert: I think that's one of the hardest things of modern machine learning in healthcare the generalizability I think you have to do really heavy cross validation experiments.

Robert: You have to, and this is why the FDA is so stringent on, did you test and validate your algorithm on a variety of people, a variety of types of patients? Were they all in the same small town that you ran your initial training set on? Maybe they have a certain thing in their water or the diet that made them different than people in the town over.

Robert: So I think the whole thing is how do we validate these AIs on larger and larger cohorts? The fewer [00:20:00] patients you evaluate them on, the less it's going to be generalizable and actually leaving a set of patients out for validation, uh, that you never touched during the product development or predictive learning or anything.

Robert: That's gonna be super important to try to make it more generalizable. If it works for Americans, will it work for Canadians? Maybe there's something in the food there that's different, all that maple syrup, you know, things. But I think just testing the generalizability on larger cohorts so that when you do the clinical trial, it's hopefully not over trained on the training data that you fed it.

Robert: And that's why a lot of the good AI researchers in medicine use foundational models where you're starting the neural network with a set of neurons and weights that were trained on a huge corpus of data, like the whole internet, or a foundational model for radiology. You might have seen a million different MRI scans from a variety of different diseases first.

Robert: And then you fine tune that to [00:21:00] your dataset. This way, it's more likely that it's going to generalize if you started with a foundational model that, let's say, Microsoft releases. So Microsoft just released a foundational model recently that combines x rays and surgical videos in one neural network. By using these foundational models, you're borrowing all the research and intelligence that a large company like Microsoft might have put into data gathering and neural network training.

Robert: Or, like I said, for organ segmentation, you start with Facebook's SAM model, because that was trained on billions of photographs. Okay, and then you adapt it to medical imaging. You start with the foundational model, you then adapt it to kidney disease for this part of the country, and then you test it on another demographic.

Robert: Set, you know, so starting with as much foundational models that were trained on a wide variety and corpus of data is Is how do you mitigate the risk that it won't work well in practice?

Mehmet: Yeah, do you think here, you know the use cases of real world data would become more You know [00:22:00] useful in this situation.

Mehmet: So we have like more large data sets

Robert: Uh, I wish I hope but everyone's so squirrely about their data sets in healthcare and no one releases it There's not enough open source data sets out there and data sharing so I think that the whole community of research and healthcare AI would benefit with more public and open data sets.

Robert: It's just so much, so much alpha, as they say, is locked in the servers of these hospitals and imaging centers and, you know, private enterprises or public enterprises that are scared of HIPAA so they don't want to anonymize their data and let researchers use it. So all this data is sitting there. Not really being used to train until a big company like Microsoft comes in and says, Hey, let us train with a million images that you have sitting on your servers.

Mehmet: Yeah. Probably unless they find a way to anonymize, anonymize the, yes.

Robert: And you can anonymize. The problem is anonymizing at scale. Most of these PACS systems and these EHR record systems, they let you [00:23:00] anonymize one at a time, not millions at once.

Mehmet: Absolutely. So you need to do that at scale. You're right.

Robert: Yeah.

Mehmet: Um, Now, jumping to more, uh, towards us, you know, the, the, the patients, okay. Hopefully everyone is healthy and they, we don't have to call them patients, but you know, we have to go to doctor and we have to do checkups also as well. Now, my question is. How are you seeing also, let's say the mass, all the people right out there are ready to accept also to be maybe diagnosed by AI, maybe, um, you know, getting a prescription, which was also suggested by AI.

Mehmet: Do you think us as humans are ready to take these or, Oh, You know, it's gonna take some time until, you know, we, we accept that the fact the machines sometimes can do, I would [00:24:00] not say better job, but maybe they can act faster than us humans because You know the nature of you know, the machine can do much more calculations I don't know like operations faster than us.

Mehmet: So how we tackle this acceptability I would say of AI in Healthcare,

Robert: so I think what we're noticing is people are starting to use something like chat GPT as a second Look, so the doctor says I have this Let me just get a second opinion, but not by another doctor, by the AI. So I think the first foot in the door for AI in people's individual lives is going to be a second opinion.

Robert: Let me do my own research. Doctors hate when you come in with WebMD stuff printed out, right? They think that, hey, you can't replace a medical degree. But I think more and more people are going to start using AI as a second look. I think people who are hesitant to go to the doctor are going to use AI at home themselves.

Robert: If the A. I. can look at their scans and [00:25:00] give them a, um, suggested prescription, people, sometimes people are more introvert and they don't want to go see a doctor and tell all their problems too. You see the A. I. therapists taking off to some degree. Some people find it easier to share. Confidential private information with a, you know, impersonal robot than another person at the other end.

Robert: So, I think there's going to be adoption there. I think there's going to be pushback when people are like, Hey, you know, I'm not going to blindly take this pill an AI recommended I take. I don't care if it's board certified, it passed the MCAT, whatever. I'm not going to swallow this pill just because an AI 3D printed it for me and says it's gonna work for me.

Robert: I want a, I want a third opinion from a human. So I think just adding an AI suggestions into the workflow of how people deal with their health is going to happen gradually with second opinions and then doctors are gonna find that they're not doing a good job if they're not also using AI to [00:26:00] second check themselves.

Robert: I mean not, no one can keep up with all the research coming out.

Mehmet: Right. Now, still within the AI, because people might say why we're talking too much AI because, you know, it is, it is the current technology, of course, but there must be like other, I would say, technologies in health tech, um, that you're seeing like great use cases combining them.

Mehmet: Like you mentioned about, for example, 3D printing, right? But are we seeing like something else? Are we seeing something related to wearables? Are we seeing something related to M. I. O. T., which is Medical Internet of Things devices? So what are like some of the trends, if I might ask, Robert, that you're seeing and you're excited also about?

Robert: So I think two main trends are Remote medicine. I think teleradiology, telepathology, telesurgery, like, just a really good strong connection where someone, a doctor, can You know, be on call across the [00:27:00] country. I think remote work is there. I think VR is bigger in medicine than people think. Imagine the surgeon goes, puts a VR glasses on and simulates the whole surgery at once beforehand, or takes a little tour inside the person's body with VR before doing the surgery, I think VR.

Robert: We'll allow people to go into a digital waiting room in the metaverse, for instance, and then talk to a doctor in the VR world. Um, I think VR was going to be used for, um, let's say, pulling up patient's blood work, their chart. There are medical imaging, all in nice interface, and that might be it's AR, you know, augmented reality, where it's like Pokemon Go, you see something pop up in front of your screen and with your glasses on, so AR and VR, I think are going to be huge in the surgery and radiology world.

Mehmet: Yeah. And, uh, you know, I don't think we are very far also from there, but, uh, I see this and I discussed it with a couple of my guests [00:28:00] before. The main benefit I see is we hear sometimes like in some countries they have shortage on, uh, medical stuff. Right. So this might fill the gap and even some like less developed places in the globe.

Mehmet: We see that even they don't have the basics, right? So this might also become one of the ways where we can fill these gaps also as well. And this is why I'm super, super excited when, when I see these breakthroughs, you know, technologies and the innovation that are coming out from there. But I want to go back a little bit, Robert, to the regulations.

Mehmet: Um, of course in the U. S. You have the HIPAA in Europe, the GDPR, and all, you know, whatever comes out of there. Um, what needs to be done? Like, I understand why these regulations came in the first place. Of course, to protect the privacy of the data, to make sure that no one can use the data in [00:29:00] doing bad things.

Mehmet: Let's put it this way, to make it simple and clear to the audience. Uh, But who do you think should lead first to start looking into these regulations? And maybe, I would not say remove them, but maybe modernize them to be, you know, at least, you know, modern enough to grasp All these advancements that we have with the AI.

Robert: I think when it comes to regulations, um, I think it's very onerous for startups and smart entrepreneurs to enter healthcare and healthcare tech because it costs millions of dollars just to get a product onto the market. That's not the case in any other industry. E commerce, you don't need millions of dollars to launch a dropshipping service, you know?

Robert: I think the regulations need to, one, be more prominent with displaying [00:30:00] how many patients an AI system was validated on. That's one thing I think we are under regulated on. If you look at it, a lot of these things are approved with a couple hundred to maybe a couple thousand patients validated on. I think that these need, that number needs to be very prominently displayed on any AI device.

Robert: You know, this was an AI model, great, FDA approved. Instead of doctors blindly trusting that, how many patients was this validated on? What was the accuracy? What, what, how often did it get false positives and false negatives? Now, I think we need to remove the onerous, uh, stranglehold of regulations on startups, entrepreneurs, because we're not innovating in medicine because It's, unless you already have a way to raise money from professional investors like VCs, it's difficult to actually get an FDA approved medical device to go to market.

Robert: And it's difficult if you don't already have a hospital you're working with. I think there needs to be pathways that is, right now it's either, hey, what I built is for research [00:31:00] purposes only, I can't market it, I can't sell it, or FDA, 510K, PMA approved. I have a whole marketing budget. I spent five million dollars on, you know, dealing with regulations and clinical trials.

Robert: It's in the market and I can be like GE and just do whatever I want within reason. There needs to be something in the middle where we are encouraging young entrepreneurs to enter the healthcare AI world with ideas without them being turned off by the onerous regulations to go to market. Maybe it's like a like preliminary, you know, not Uh, research only not full FDA approved, but like a fast track on track to get released to the market where, you know, there's extremely heavy limitations on what you can say, but you don't need to spend millions of dollars on clinical trials to get to market.

Robert: So the regulations right now. Is, are good because they're trying to keep the product safe. They're trying to make sure the AI wasn't just trained on one subpopulation and doesn't work well on another [00:32:00] subpopulation. So that's what they're good at, but they're bad at giving away for budding entrepreneurs to want to innovate in the medtech space without needing to do a full, onerous, high reporting, regulated product to market.

Mehmet: And I think there will be also the ethical aspects, I believe, also, Robert, which people sometimes, you know, they say, okay, fine. We can, we can accept that maybe we need to ease the regulations. But what if someone tries to do something bad? Of course, I don't agree. But I'm taking the devil advocate role here a little bit, if you want.

Mehmet: So how we can defend this?

Robert: Well, I think it's like, okay, so, you know, John and Jane Smith start a company like Theranos or whatever and they do something and they didn't test it on enough patients and now they're in the market and a patient dies because they followed the recommendations of something that the FDA did not regulate [00:33:00] enough.

Robert: I think you, how do you, how do you mitigate against that is Making it very visible that this is preliminary. This AI was not trained on a sufficient number of patients. Yes, you can still make money on it and you can buy a service, but like, you know how like pharmaceuticals have to list a thousand side effects for any drug they put to market?

Robert: Make it so scary, but let them at least go to market. So in terms of the ethics, make the, any company that goes to market, Even with less regulations, you have to report and make it really visible. This might not work for you. You know how ChatGPD, you know, we can potentially spew misleading information, whatever it says.

Robert: All these disclaimers, this is AI generated, don't trust your life with this. Still let them go to market though. But, like, stamp on a thousand disclaimers of, scare the patients away from using it.

Mehmet: Right. That helped

Robert: mitigate that.

Mehmet: Yeah. And you know what? I tell people sometimes, especially on the content, right?

Mehmet: So, and I think, you know, [00:34:00] last year or two years ago, it's been two years now. So one of the things that we discussed on the show, it was with Sid Mohasib and hi to him if he listens. So. When it comes to finding content or information, especially with healthcare, I've seen it a lot, and I used to do it, and I still do it, honestly, sometimes.

Mehmet: Like, if I have a pain, let's say I have a pain in my elbow, and I go to Google, the first result I start, and of course, you know, like, sometimes we don't read this, and we know, like, sometimes some websites, they would have their SEO in a way, so They put this keywords and they just put something nonsense, right?

Mehmet: And the AI is not different because the AI actually is generating a content based on the knowledge base that it has. Of course, it's not like the search, it's doing kind of retrieval over there. So this is why, yeah, the ethical aspects, I agree with you, Robert, here. And this is why I was kind of playing the, you know, uh, devil advocate here.

Mehmet: But 100 percent on what you're saying here, Robert. Now you [00:35:00] mentioned, you mentioned entrepreneurs and startups and Part of, you know, what I do with the show is always asking my guests what we need, you know, to do, to have like more founders and entrepreneurs in healthcare. But other than the things that you mentioned, like what other advice, you know, you would give them based on your experience.

Mehmet: They've been doing this for a long time yourself. So what they should think about, what are the, you know, kind of best practices, maybe things that they should do, things should avoid, of course, on a high level. If you can share that with us.

Robert: Yeah. So, you know, I've been, uh, an entrepreneur in the healthcare AI space now for 11 years.

Robert: And, uh, my company is data tech dot AI. If anyone wants to check it out, we are custom healthcare AI consulting. Um, and I think for founders, you got to start with the data. In healthcare AI, you don't always want to start with an idea. That could be great. You got to start with what data is available. A lot of the idea is you're [00:36:00] gonna be roadblocked by not gathering enough data to train a good AI model.

Robert: Um, start with open source data set. Start with a doctor collaborator that can give you some preliminary data. Start with a university connection or research lab that has some data you can work with. Uh, scour the internet for public data sets. I think healthcare, ai budding people, they should look on hugging face for the latest AI neural networks being done in healthcare, but then look for the data sets.

Robert: There's data sets of doctor patient conversations that you can use to train an LLM or fine tune one. There's open source, uh, radiology data sets. There is surgical video data sets. So, whatever an entrepreneur does, one, enter it knowing you're going to have to deal with the FDA. Secondly, start with the data.

Robert: Everything makes it easier when you're, when you're starting your endeavors with more data rather than a clever idea. That's what I would say.

Mehmet: Great, Robert. Uh, as we are almost coming [00:37:00] to, to the end, you know, and this is usually how the final question I ask any, you know, maybe final thoughts you want to share with us.

Mehmet: Maybe something I didn't touch and you want to just quickly, uh, say something on it. And of course, where people can get in touch.

Robert: Yeah, so, um, you can get in touch, uh, go to thetatech. ai, there's a contact form, and then it'll go right to my email, so thetatech. ai is where to contact me. Um, in terms of things we didn't touch on, I think it's really exciting how cool and futuristic all this is going to look.

Robert: When we go walk into a hospital, it's easy to get scared of AI replacing doctors or AI giving bad judgment, or under regulated or over regulated, whatever. But you're going to walk into a hospital, and it's going to be some cool Star Trek stuff there. It's going to diagnose you more accurately than a regular doctor, maybe not as accurately as a specialist.

Robert: We're going to have custom, personalized, Cancer treatments based on your DNA, [00:38:00] your blood work, your Fitbit monitor, whatever it is. I think just the futuristic, cool stuff coming out of it is going to be really nice. And also there's just going to be a huge shift to patients owning their own data. I think there's going to be a lot of budding entrepreneurs and startups that are going to let you not rely on keeping the data in the hospital's data centers.

Robert: But you own all your data. You can easily have an AI locally with an open source AI model, uh, analyze it. You can give it to another doctor easily. I think there's gonna be a push for consumers to, to, to collect and own their all data. And you always see these VCs funding companies that say they're going to do that, but AI is going to let it actually happen.

Mehmet: That's, you know, exciting times I had. I can see this. I can feel it also as well. And, uh, to your point, you know, I'm waiting to see the impact of AI. Of course, we started to see already. Don't get me wrong. But, you know, more impact, more, uh, life [00:39:00] saved, more, you know, uh, You know healthy lives I would say in general so people doesn't have to suffer Uh because of the technology that we have in our hands and you know and kudos to everyone like including yourself robert like Whether it's on the research part whether on doing practical and building Ai models and helping you know organizations and healthcare institutions to reach that step.

Mehmet: So You know, all the thanks goes to you and everyone who tries to push, you know, the boundaries over there Um again, you know as I said like this was a great fantastic conversation with you today for the audience The link that robert mentioned you will find it in the show notes. So You don't need to search much So you'll find it in the show notes.

Mehmet: If you're listening on your favorite podcasting app, or if you're watching on YouTube, you will find it in the description. And, uh, as I say, always, this is for the audience. Thank you very much for, you know, uh, all the support, all the, you know, [00:40:00] comments. You did something great this year for us on the podcast, on the podcast, you managed to bring us up in the top 200 charts in multiple countries at the same time, which is the first since two years.

Mehmet: The second thing you put us in the top 40 business podcasts in Dubai ranked at the 14th. Thank you so much for that. And if you are just new to the show, thank you for passing by. I hope you enjoyed it. If you did, so give us a thumb up, share it with your friends and colleagues. And as I say, always, thank you very much for tuning in and we'll meet again very soon.

Mehmet: Thank you. Bye

bye.