Revolutionizing Pediatric Care with AI: Grand Rounds Talk!

Music. So the mission of matrix ai consortium is really to think about inventing deploying and advancing ai at scale with the primary goal of building technologies that advance or promote human well-being and quite often when we say human well-being health is of prime importance but But in general, it could be any aspect of human well-being that could be boosted or accelerated with these technologies.

And in order to accomplish this, we really take a transdisciplinary approach. So in the consortium, we have AI researchers, ML engineers. Music.

How humans process information, how we make predictions, how we learn in different environments, and how we do that in the most efficient way. But most importantly, when we are deploying any of these technologies, we have to think about whether the solutions we are building are trustworthy. Music. Whether the solutions are robust, fair, unbiased, whether the models are interpretable. So we have a trust on trustworthy AI. And ethics is a cross-cutting theme across all of these.

And today, the consortium has centers of excellence in each of these trusts, supported by different agencies. So we have about $45 million in funding portfolio with 30-plus research labs and, as I said, five centers and students engaged from multiple grad programs. And about 500 students from different research labs were part of the consortium. And we have AI compute infrastructure.

I think this is a modest infrastructure if you think of the resources that some of the computational researchers would have. We have DGX servers, Lambda servers, and we have a very niche infrastructure that we recently got funding from National Science Foundation. We are the national hub for. Music.

Okay, so that's kind of a brief overview of Matrix, and I'm happy to have conversations offline about any interests about what Matrix is doing and how to engage and so forth. So for the rest of the talk today, I'm going to focus on kind of where AI is today in a very specific aspect of how algorithms are doing today and what are some of the more recent advancements and some opportunities specifically in pediatric care.

So if you look at the trends that are out there, I think this is publication or data points from Harvard School for Public Health, it shows that there is a projected reduction in treatment costs when using AI for diagnosis. You could get like a 50% reduction in treatment costs. That is a significant and reduction, high cost efficiency. So why then not consider this in the loop for diagnosis? And similarly, there is a projected improvement in health outcomes up to 40%.

I'm sure there are a lot of caveats to these projections, but in general, there seems to be a huge opportunity in achieving some savings by using these tools. And there is an estimated market price that you would see for healthcare and that is a significant investment that is coming in. So it's really, I would say these are not ordinary trends and I think this is only going to increase in the coming years.

So it's very important to think about AI not as a passive tool, but how do we actually actively engage in the diagnosis or in the prediction or in your workflow, how to actively use it. So when I say passively using it as a tool, I am sure some of you are using ChatGPT or some form of large language model today and maybe querying it in some forms, but is it part of your workflow where it can help you in more active ways? I think most clinicians are not doing that at this stage.

And I'm sure for several reasons, beyond the technical flows that are not designed to support it, it's also infrastructure support and so forth. But it's very important to think in this day and age to. Actively use this in the design flow. So at Matrix, we kind of look at different projects that are focused on AI centers of excellence that are related to health.

I'll give an example of one of it in the later slides, which is the chatbot we are building for the trauma research supported by the state where we are collecting data, trauma data for the entire state of Texas. And there are a few other projects where we are trying to build tools for physicians and clinicians in adopting AI in their workflows through AimAhead and other projects. But I just want to give you a context in terms of the hype that is there about AI.

So several of you might be familiar with Mark Roberth's videos. If you look at this video in here, you see that this animal is able to really adapt to the changes in its environment, come up with creative solutions to get to the destination, and it is continuously improving on how it is doing this. This is a complex environment. Whereas if you think of modern day AI, it is not able to adapt in this form. This kind of flexibility and real-time adaptation for different tasks is not there.

So if you think of current AI models, whether they're AI agents or not, they're really incapable of such continual learning or adapting to unforeseen challenges in real time. And a lot of times there are huge data sets that are needed to solve these problems. So we did this project with DARPA. This is almost like five years or four years old now, where we were trying to solve this problem of how do you build AI models that learn continually in different environments.

So I hope you're able to see this video. So we built these new types of algorithms that are able to adapt, and this is deployed on this drone in a photorealistic simulation where it is able to forage through this dynamic environment and try to reach to the people who need rescue operation. So in search and rescue operations, you can perhaps consider these kinds of algorithms. If you look at the current ones, this kind of adaptability is not there. And of course, this has to be done in...

In a matter of milliseconds. So which means latency is key aspect as well, not just adaptability, but how quickly it is able to do it. So these were some early attempts in addressing this problem. And since then, we have advanced the algorithms quite a bit in our work. And there is a lot going on in the community in order to think about how do we design AI that learns continually similar to humans, a lifelong learner.

So for the next part of this talk, I'm going to show some examples in settings that might be familiar to you in how AI is being used or can be used and potentially used. So very early on, I think one project that we worked on was with the city of San Antonio, where during the peak of COVID, they wanted support on trying to see if AI could help them in understanding the patterns of, you know, the mobility patterns, sorry, in the city.

And based on that, if we can predict how the COVID spread is going to be impacted within the And these were some things we have done on almost a weekly basis early on, and then it almost became a daily report that we gave to the city of San Antonio. These models were built by graduate students in Matrix to make those predictions. And like, you know, if you increase the mobility by this factor, what would be the effect if you increase by 50 percent, 75 percent and so forth?

So this was, I think, helpful for the city at that point of time to have a real-time dashboard and have a model that could make these predictions. So if there are situations where real-time decision is needed with different types of data in your workflows, I think this is something Matrix can support it. The other thing we have done more in advanced technology is like, you know, quite often, well, especially during COVID-19, you have these different strains that have evolved, right?

Like you have the alpha strain, the beta strain, the gamma, and so forth. What happens is if you design an algorithm, like when we designed it, The algorithm was looking at certain strains, but a new strain might have other features that need to be embedded in the algorithm. So we have started looking at the continual learning algorithms that could help in predicting this kind of changes without forgetting what it has learned in the past.

Because a lot of algorithms, they're very good on solving the current task, but they forget what they have learned in the past. And this is called catastrophic forgetting or catastrophic interference problem in machine learning community.

So we're looking at algorithms that would solve this, and this is a great case study for it in how you can build dynamic architectures so that when the strains or when these viruses are evolving, you could use more modern architectures to solve that problem without losing the information you have learned in the past, similar to how humans do it. And if we look at other ways, if you look at this cartoon, I think there are different sources of information from which you can.

Learn and try to get a full picture of a patient's health. So here you see the diagnosis codes from the doctors or you could have textual notes that come in from the clinical data or there is imaging data of different modalities. There is cell imaging data and there is also these nodes that are written, or we would call these unstructured data. And you can think of other, you know, EG, ECG, you can have claims data, you can have genomics data for this patient.

So these are what we call multimodal data, as in like, you know, the data is in different formats. It's coming from different sources, and you use this information and fuse it with multiple algorithms to understand or diagnose the patient's health. And for example, what I'm going to show in the next slides or what you might be using, something called ChatGPT, is using one modality or two modalities of information. I guess it's using three right now. It's using text, voice, and image data.

So, if you have this multi-modal input, the algorithms become a lot more challenging in order to get the right output. The algorithms don't get challenging. It gets challenging to design the algorithms. It's also important to understand where the data is coming from and if the data is being fused in the right format to make the right decision. So the sources of all these data formats becomes very important.

So we kind of did this single modality one in the trauma project that we are working with the state, which is TRC4 funding for this. So Dr. Khutub from Matrix is leading this project. So you can see that. Let me play this video. I hope you're able to see it.

So a trauma assistant was built where it gives some more nuanced information of what action should be done in an emergency situation. And it runs in like, you know, a matter of milliseconds. So here is an information. A 28-year-old male falls from a height of 10 feet and presents with numbness and weakness in both legs. and he has normal sensation in his upper limbs but is unable to move his legs and has absent patalar reflexes.

So what action should you take and does this patient require a level 1 trauma center? So when we run that, I hope the videos... So it is giving guidance on the actions that could be taken, like, you know, immediate actions, whether there's immobilization, assessment, vital signs, monitoring, establishing IV access, and then referral through the trauma centers and the level of care that is needed.

So here in summary, ensured rapid transport to a level one trauma center is initiated while monitoring and stabilizing the patient as per guidelines.

But then you can think and imagine further where it gives like dynamic information on what is the trauma center that could be, I mean, though the, I think, ground level crew would know the nearest trauma center, I think you could have access to the usage, the wait times, everything dynamically in a visual so that you, and the traffic that you would have all these kinds of information in a single

flow where you could be guided to go to the right trauma center and get the care immediately. So. This kind of application is possible where you have here a singular modality of information. This is not multimodal, but multiple other inputs can be given to this assistant. And that's what this was a pilot study. And we are advancing this with now collecting information of all the data from the state of Texas. And Amina is looking at advancing this chatbot to do a lot of these additional features.

That would support care in these situations.

The other tool I thought you should know that got released literally last week or 10 days back, it is called MedGamma. It is again a language model but it's particularly trained with medical data to support diagnosis or a lot of pre-analysis and care. So I wanted to show some of the demos for the MetChemma because I think if you want to explore or just, you know, even play with and see what it is capable of, this would be a good tool to explore. Sorry.

So if you have a use case, you can define the use case. Let's say you want to generate a text report and you want to get an evaluation, but you want to get performance at a certain level. Like, you know, do you want, you can set the performance that you would want, whether you want a high performance or a low performance. And I'll tell you in a second why. That would change the outcomes in some cases. But so you could look at the data. You have the input in here with multiple images.

And then you're looking at certain cases and trying to see what, based on your setting of what the performance is, you get certain response. So the reason you're tuning that performance is that you're able to process it faster on a lighter model. And this is particularly important when you're thinking about embedding these kinds of tools in your workflows because they consume a lot of energy and which means a lot of money.

It costs a lot because of the compute resources they need. So you would want to reduce that burden. And one way is to kind of use lighter models. And if, let's say, 99% accuracy is not needed in a certain situation, maybe 90% gives you the confidence to go in a certain route, I think you could change those or adjust those performance metrics in this model. So it made a big splash in the last one week. It's a 4B model is what they call it. So I would encourage you to explore it.

And they kind of show it how, like, you know, they demonstrate it, like, you know, how it looks like for an intake optimization with this tool. So there is a built-in agent that is going to ask questions to gather information about the patient. So it kind of generates an intelligent pre-visit report that would help you as a provider or a caregiver and give you a detailed evaluation of the pre-visit report.

So this is just a simulation of how it runs with what it would ask the end user, so what they're visiting for and based on that, I think what they feed in and the questions or what the queries it gives to the patient, it kind of gives you a generic summary. And you can see that some details that are not mentioned and all of this additional information that would help you to diagnose or help you to provide the right care for this patient.

So obviously, I'm sure you have lots of questions on whether the output that this tool is providing is reliable, but I think this is where collaboration with humans, experts, is very critical. Because based on the data, the generalized databases that it is looking from, the outputs that it could view sometimes may have mismatches, may have incorrect information. And that is where the expert is able to quickly say, well, that, for example, those triggers do not fit with this portfolio.

And you would call it out and the model is able to adapt. So I would say this can be used as a secondary or a surrogate still because of some of these concerns on how the data is being processed, the source of the data. But I think compared to several other models that are out there, this model seemed to be doing really well with medical data.

Some other examples where you can think of where AI could support is, let's say, a lot of models that are built today or a lot of analysis that is done is done on adult medical data, not as much, let's say, pediatric data sets are not there for a lot of diseases.

So what I mean of course it would help to evaluate on pediatric data sets and get the right diagnosis but there are also AI models that could transfer the knowledge that is there from the adult data sets and try to make diagnosis with pediatric data so this is again something that is still in the works and a lot of models, for example, what we have seen just now with MedGamma, one of the models, the large language models, there is a lot of personalization and,

But there is also opportunity to translate complex information with a language that patients can easily process or understand.

So the tone, the terms, the jargon, all could be reduced when you're communicating with kids. And also there are tools, including ones we are building on, that provide real-time monitoring and decision-making, And also there are researchers in Matrix who are looking at how do we ensure that what we are building is secure and sensitive information is protected.

So, for example, in the case of trying to predict the variations, you could think of AI that is processing vast amounts of data, basically. And it's pinpointing subtle signs of a disease that might otherwise go unnoticed. So this will help in early intervention and all of that. But well, what would you do if you didn't have the data for the specific disease you're looking at or if it is really scarce and difficult to collect data for that specific situation?

So then it's really challenging for the model to make the prediction. So there are newer techniques like transfer learning and the continual learning models. One of the features is transfer learning that could leverage existing data sets and adapt this knowledge to the unique situation or characteristics that you were looking at and try to give a better diagnosis or diagnosis or so forth.

And similarly, the large language models where I said you could have an opportunity to think about how to generate information that is more digestible to the end audience. In this case, you could have the model, let's say, if you look at the example on the right here, you see that there is a specific information about this condition that was provided by the doctor. And you could translate that in a different way to an adult and a different way to the kid.

And I think you do that on real time on a daily basis. And you could have the models also do this, not in real time, perhaps before, during the entry-free intake time or post-visit care. In different situations, you could think of using this type of individualized and age-appropriate or customized responses. And the other example I was saying was where you could do the real-time monitoring and prediction.

There might be situations, I think, there are smart health homes is like a classic example where we can think of using these kinds of sensors that monitor vital signs and use that information to make a certain prediction if there is an anomaly in the signal.

And you can think of using this in different contexts not just within a smart health home and on a regular basis like you know you have all of us have some kind of smart device now monitoring our critical signals but then fusing them in trying to make the predictions i think you need really a light device that is able to do in again a matter of less than i think 10 milliseconds. So the neuromorphic computing technologies, we are building help in this kind of analysis and prediction.

And the last one I want to highlight is about privacy. So a lot of times these data sets, whether those are proprietary data sets that. Are designed in your clinical setting or something that you're using out of a general database, you might want to secure this data, And I know there are certain standards that you follow with the HIPAA compliance and all of that.

But in general, instead of having a single cloud infrastructure or a server where you store all of this data, you would want to think about something called a federated model in how this data is being saved and how you learn from this data sets that are stored in this cloud. So what a federated learning infrastructure does is there is distribution of information across multiple clouds, many clouds, fog. There are a lot of these terms that are used in the compute community.

So you could have that distributed and you're just not learning from a single data source, but then you're learning from these, different data sources and making the decision, whether it is diagnosis or treatment or so forth. And this is something that is commonly used nowadays when you're trying to chat or messaging or in any of those tools, federated learning is naturally embedded in your smartphones or devices.

And these could be extended to more of the medical data to ensure that the data is more secure. So the end user only can see data that is relevant to them. They do not have the visibility of the generic or other data. So it is more secure in this using this federated learning approaches. Sorry, I think this is, again, an extension of another example of MedGammer that I wanted to show in diagnostic support. So you can see the prompts that are given here.

Again, if you're an expert radiologist, please describe the findings for the bow chest X-ray. And you see the radiologist impression on the left. And you could also see on the prompt on the right, what do you think is the most likely cause in this image?

And what would be the next steps that you would suggest and this is like you know more a quick response to see like well consult a dermatologist for this additional steps so you can think of using this in different ways in your workflows it could you know integrate multiple images again here both 2D and 3D scans it can generate comprehensive and reports and there is visual question answering is what you see on the right.

It's called the VQA technology where you're giving that image and asking it to analyze it and you can query about the image. And you can also think about using it in clinical decision support. Again, this is a caveat. It comes with a big caveat, right? You want to really think about the data sources and inbuilt models and the level of accuracy this model is giving.

But since it has just been recently released, I would look at some of the example scenarios and try to work with it and see how it does without giving any proprietary data. And similarly, I think I've already shown this, so I'm going to skip that.

So as you are thinking about adopting or integrating more actively any of these models in your workflows, there are multiple things to think about. Well, the model does great in terms of performance, but does it work in all the situations? Is the model fair? And why would this happen? Why would the model be not fair? The reason is it might be trained only on a subset of the data. It might be trained only in certain geographical regions.

Maybe it does not translate well across the country in different scenarios, different geographical regions. So it's very important to be aware what was the training data for the model and what are its limitations before using a model. So that it doesn't perpetuate incorrect information or it might not amplify unfair predictions in some scenarios.

Similarly, with the output that is generated from this tool, if the data set it is relying on is not a robust one, it can generate incorrect information. And this is where human AI teaming and collaboration is very important to correct the outputs of this model. As an expert, you're able to do that. So it's important to not fully rely on the outputs that are viewed from the models.

And I think this is probably more of a concern in human-rich environments, and this is a. Or in mission-critical environments where if you're using AI, you have to think about the accountability. And I'm sure you are a lot more cognizant and trying to embed this in your workflows more actively as to how to reduce any kind of compliance issues.

So when you're using these types of models, what if the model has predicted something wrong and you have relied on that information and has given the treatment based on that? So how do you ensure that this kind of impact is reduced? use. So it's very important, again, to tie to the earlier things, whether the model is fair, whether it is giving the right information, whether it is trained on the right data sets. Is the model matured enough to make accurate predictions?

So this kind of information is helpful in deciding whether to use that model because of these kinds of issues on accountability and so forth. And the other point to think about is whether the model is even trained on open datasets or are there going to be some IP ownership issues on where the model is getting its data from or where it was originally trained on. There are models which are originally trained on and then you just deploy them in your scenario.

They don't need that additional training, but there are other models that are continuously training and updating based on the information that you're given. So in either of these scenarios, it's important for you to think about where that data is coming from. So I guess you see that underlying theme for all of these is to be knowledgeable about the data that the model is trained on and the source of the data and whether the reliability of the data so forth.

So if you're more interested in trying to understand some of these, we do have sessions in tinkering with an AI scientist or we do bootcamps for MD students in MDMS and AI program. We were doing bootcamps earlier. So if you're interested, we would keep you in the loop when the next training session happens.

So thank you so much for your time. And I'm open to questions. Thank you, Dr. Kuditipudi, for that wonderful presentation on transforming healthcare with AI. Let's see if anybody has any questions. So let me start with me. So most of us are sort of scared of AI. We have not used it. Where do we start as a novice in AI in the medical field? Is the MedGamma a good place to start?

It's a good place to play. What I would say is, I think no matter which profession we are in, everybody should be an AI generalist in this day and age. So what does that mean is to be aware of some of the tools and actually try to, it takes very minimal effort in playing with these tools because the user interfaces have improved significantly. Definitely, I think large language models is something I think that is, if not today, pretty soon going to be embedded in your workflows and thinking.

So it's a great tool for you to play with. And Medgamer is, yes, definitely something I would play with. And I think you can immediately at least see, oh, well, it does really well in the situation, but it does terrible here. But you're giving the guidance to have vector models that would help you. But also, I think in general, trying to understand the basics of how these models work. You don't have to understand the full intricacy of the exact models.

But I think, as I said, being an AI gendronist, trying to understand, well, what does the pipeline look like? As in how the data is being taken into the model, as in what kind of pre-processing is done, how it is fed into the model. And then once it is fed into the model, what kind of outputs the model is giving and how reliable are they?

So I think from the AI developers itself, there are things that we should be doing in reporting these statistics on the situations in which the model should be used and should not be used. But I think to answer your question, I think it's really important for everyone to be an AI generalist. And this is a good tool to start with. I was actually going to compile a list of tools that would help you. And I can share this after the talk and if anybody is interested in playing with a series of tools.

So there's a question in the comment box by Dr. Lip said, very interesting, thank you. Will AI eventually completely replace the physician? I think this is where the hype is, I would say, right? I mean, yes, AI is able to do some of the tasks that are very streamlined or, you know, where the data is streamlined, where the data is structured, where the environment is structured.

It doesn't do very well in unstructured environment still. It doesn't learn continually or adapt as well as humans do in different situations. We are far from that. There are a lot of claims that, you know, today we have an AGI is going to come, which is artificial general intelligence, or you might hear super intelligence. You hear all these terms that are thrown out there that it's going to happen in the next couple of years.

And a lot of researchers who are in the field do not think it's there. Because of several underlying fundamental problems that need to be solved with these models. And again, I think we humans, we are 3D structures, right? So we move in environments and we do a lot of things beyond just processing this textual or information from the images.

So if you think of robotics or other tools that are going to support the clinicians or physicians, they are also they also need quite a bit of evolution to even think of working in these unstructured environments they work very well in factory settings where everything is streamlined.

So I personally do not think that they are going to replace the physician but there is going to be a lot of teaming up that's going to happen. So this is why I would encourage everyone to be an AI generalist. Thank you. This is Dr. Hedgin Beller. Do you have a question? I see you have raised your hand. Yes. Sorry, Dr. Kumar. This is Dr. Asanas, and I'm here with my residents on the inpatient service. Dr. Hedgin Beller was actually the one that logged in for us.

So again, my name is Dr. Achachwan Asanas, and I'm a pediatric hematologist oncologist, and I also work for the physician's practice as an administrator. You mentioned the issues with With regard to HIPAA and utilizing the federated model, there are currently AI platforms in use for physicians kind of collecting data with regard to the interview process. I think I can mention a bridge here. They have mentioned that they erase the data after a period of time.

As AI generalists, what should we be looking for? really the data model of how to make sure that we're using secure mechanisms? Or should we feel fairly comfortable when people say that they're erasing the information before anybody outside may be able to access it? Very good question. So, I mean, when someone is saying they're erasing data, well, I think perhaps, I mean, the way they designed it, perhaps it is like a three-month time frame, six-month time frame, the data is erased.

But what happens is it really depends on the underlying architecture. It happened with some of the, several of you might be using messaging platforms like WhatsApp. I don't know if you're familiar with it. There are some of these messaging platforms where the user was thinking or informed that data is being erased. But in actuality, because of how architectures are built behind where the models are deployed.

The data could be residing in some or some artifacts of this could be residing in some form in one of these architectural building blocks. So how do we know as an end user? Sometimes this is really only experts can see that. And sometimes these things come out, you know, much later than you would like to know about. So what, I mean, I'm not trying to scare you here, but I think it's very important to ask questions about the sources of data.

And the architecture they're using and the confidence rate at which they think this data is going to be, data is being erased and ask them to show you the flow, how data is being erased. I mean, probe on it further to kind of see what is the process. And what I would also ask is like, consult with an AI specialist, not just with a vendor you are working with, but consult with an AI specialist.

I mean, there are several that we could refer to from the matrix consortium, for example, who may not be expert in the exact technology you're using, but they could help you ask the right questions also. So I would use that layer approach, trust, but I think be cautious. Thank you, Dirish. And then there are comments by Dr. Jones. One HIPAA, let's see. I can go ahead and comment, Dr. Kamad. I think, you know, interesting.

I presume you guys know about open evidence, again, out by Mayo and in collaboration with New England Journal of Medicine, and now actually JAMA Network as a tool. It's free and available to clinicians. You have to have an MPI number to use. And it's HIPAA compliant now as well. And I actually just ran your question through on the trauma patient that you had. And he actually gave a rather remarkable answer, very thorough with the references that you can then tap on.

So open evidence is actually used. Now, again, you talk about a place to start. I warn the residents routinely when they use these tools, prompt engineering is still very, very important, and particularly in pediatrics. If you miss a key element, and age is one of them, to put into these tools, you will get very convincing wrong answers.

And so that's still one of the biggest concerns I have as we begin to introduce this to our residents is if you don't have that knowledge and that background to see or to, when you get that wrong answer to at least turn your head and say, man, that doesn't sound quite right. You're going to miss it because it, again, it's so confidently wrong that you will make errors. And so prompt engineering, I think is, is such a key element to begin with when

you train people how to use these, these tools. Right. Right. There are bots that are being generated to help with the prompt engineering as well. Hopefully those are made more available for different platforms pretty soon because I completely agree with you. Missing keywords can give you completely different answers or inaccurate information that shouldn't be relied on.

And the other thing about references, sorry, to just add to that, earlier tools, like at least one year or six months back, it would give you references. So you would think that, well, this information is reliable. A lot of times those references are not existing publications. It would make up references with high confidence. So it's very important to check the references that are being pointed to.

I agree. I ran into it as an editor, actually. I always check the references because those are not right. Anyway, there's a question by Dr. Rizvi. How good is the news of AI hospital in China? How good is the news of AI hospital in China? I don't know if I read this specific news article or information, But I know there is a lot more active adoption of tools at all levels in some countries.

So I think there is perhaps some significant amount of training data that you would have these models being trained on. And over time, they would become stronger and more accurate. But I think there is a lot more active adoption of tools. Yeah, I think there are several aspects within a hospital that you could use smart technologies and which I could see happening in the U.S. as well.

But it wouldn't be a complete AI hospital, as I said. I think the human element is still quite significant in here. Thank you. Any other questions, comments for Dr. Kuritipati? I have one other. Again, I think one of our biggest challenges is I've been meeting with Dr. Sankri and also our CIO here or vice president for information technology here at UTL San Antonio is a lack of a ready access to an enterprise or private level LLM.

And we've been trying to use collators, as some people will label them, like u.com and other ways to improve efficiency. But we're still limited, even though we have a very powerful list of 23-plus LLMs that we can use. They're not enterprise. They're not private. And so it really limits what we can load up into our sources or what people call a RAG or interpretation or what we can ask as far as questions and data we can share.

I mean, do you see things at your level coming out of the UT system or in our partnership with UTSA, which will are merging with UTSA, which will move us towards having enterprise or private level LLMs for doing some of our work? Because it is limiting our office in the office of DME with moving forward. And I'm watching Stanford and other people I talk to all the time.

I tell them what we're doing. They take and run it, and they're already ahead of us within months because we just seem to be limited on the resources side of this equation with AI.

Yeah, I think there are two very good points to think about here. Regarding the private instance of LLM, I'm actually having a meeting with the CIO next week, with your CIO. So we are trying to think about that structure and facilitating tools for physicians and researchers in UT Health. So I'm sure this is of prime importance, and the VPR's office is already thinking about it. So let's just build this team to think about how do we help researchers.

The other part is think about the data that you have or the assets that you have within your groups. I personally believe in open sourcing, you know, whatever models we are building and everything. But there are valuable assets that you have within your units, whether it is data that is collected from the patients or even in the intake forms. There's several things, right, that are really valuable assets. So if you can think of what are the digital assets that you would have within

your units, and I would not give it away for free. to any entity. I would really think about it. I know Stanford, I think probably you're talking about their research center on health. They're being very smart and pretty aggressive in collecting data from multiple states and trying to deploy their models. I know we could do better and we could move faster.

But independent of which group is using these assets i would think for you as a clinician in your or if you're a chair of a department what are the assets that you have within your unit it may not always be a cost associated with you giving those assets but there could be other negotiations that could happen we're not at least in academic settings from the engineering side we are not generally trained to think that way but i think it's very important to add value i mean

get value out of those assets that you would have thank you one last question a question can we get an ai tool to automate billing so billing for the insurance companies or the patients can this be automated i think there are multiple tools. I don't know how it fits within your workflow, but I think this aspect of the problem is a lot more, I wouldn't say easier, but I think it is being adopted in multiple industries already. So there are tools to do it.

OCR technology has matured so much, mature recognition technology, even if it is all like, you know, legacy paperwork that you have, it could all be screened and fed into models to automate the process. So I think there are tools to look at. Thanks for listening to that informative Grand Rounds talk. I'm Holly Wehmint, and thanks for listening to Pediatrics Now, our way. Music.