Artificial Intelligence and the Jaw

For nearly a hundred years, science fiction stories have been giving us an idea of what living with artificial intelligence might be like. But we don't have to look to our favorite sci-fi to see artificial intelligence, also called AI, in action. It's already making an impact in our everyday lives whether we realize it or not. When you ask Alexa or Siri a question, unlock your phone using face recognition, or get a notice from your bank about possible fraudulent activity on your account, AI is working in the background to offer us an opportunity or information that we didn't have before. AI uses computers and machines to solve problems and make decisions in the same way human minds do, faster and often with more accuracy. This offers incredible opportunities in biomedicine, where AI can not only help us understand more about how the human body works, it can help us discover the best ways to treat patients, leading to better outcomes.

In this episode of Science Never Sleeps, we're joined by Dr. Hai Yao, a professor of oral health sciences in the College of Dental Medicine at the Medical University of South Carolina and associate department chair for the Clemson-MUSC Bioengineering Program. He also serves as the Ernest R. Norville Endowed Chair and professor of bioengineering at Clemson University. His research studies tempera mandibular joint function and disorders, also called TMJ, and why risk factors for this issue impact treatment and prevention. The TMJ makes it possible to move the lower jaw, which is important for eating and speaking.

We are also joined by Shuchun Sun, who at the time of recording, was a senior PhD engineering student in Dr. Yao's lab, studying machine learning and biomechanics. He is currently a research associate in the Clemson-MUSC Bioengineering Program.

Read The Transcript

[00:00:06] Gwen Bouchie: From the Medical University of South Carolina, this is Science Never Sleeps, a show that explores the science, the people, and the stories behind the scenes of biomedical research happening at MUSC. I'm your host, Gwen Bouchie.

For nearly a hundred years, science fiction stories have been giving us an idea of what living with artificial intelligence might be like. But we don't have to look to our favorite sci-fi to see artificial intelligence, also called AI, in action. It's already making an impact in our everyday lives whether we realize it or not. When you ask Alexa or Siri a question, unlock your phone using face recognition, or get a notice from your bank about possible fraudulent activity on your account, AI is working in the background to offer us an opportunity or information that we didn't have before. AI uses computers and machines to solve problems and make decisions in the same way human minds do, faster and often with more accuracy. This offers incredible opportunities in biomedicine, where AI can not only help us understand more about how the human body works, it can help us discover the best ways to treat patients, leading to better outcomes.

Our guests in this episode are researchers in this exciting biomedical engineering space and are using AI in their work to improve lives. Dr. Hai Yao is a professor of oral health sciences in the College of Dental Medicine at the Medical University of South Carolina and is the associate department chair for the Clemson MUSC Bioengineering Program. He also serves as the Ernest R. Norville Endowed Chair and professor of bioengineering at Clemson University. His research studies tempera mandibular joint function and disorders, also called TMJ, and why risk factors for this issue impact treatment and prevention. The TMJ makes it possible to move the lower jaw, which is important for eating and speaking. Shuchun Sun is a senior PhD engineering student in Dr. Yao's lab, studying machine learning and biomechanics. Stay with us.

[00:02:23] Bouchie: Dr. Yao and Shuchun, thank you so much for joining me on Science Never Sleeps.

[00:02:27] Hai Yao, Ph.D.: Thank you, Gwen, and we're happy to be here.

[00:02:32] Shuchun Sun: It's great to be here.

[00:02:32] Bouchie: So in our intro, we talked very briefly about what artificial intelligence or AI is, but what, Dr. Yao, how would you explain what AI is?

[00:02:46] Yao: So AI, also we call that artificial intelligence - so this is a kind of branch of the computer in science and it focuses on developing a computer program basically, so it can think of behavior with human level intelligence. So that's how we defined the AI or the artificial intelligence. So AI actually is right now, it's everywhere and so, for example, I think we all have this kind of experience. So if we... go to the Amazon and on the YouTube to watch the videos. And so, you will see the apps can really know what you like and put the kind of content you like to see in front of you. So that is amazing, how these kinds of apps can do these kinds of things.

So actually, yes, they are using the AI models to predict your behaviors. And also other examples, for example, you are using your cell phone - iPhone, or Android phone - you're using the voice assistant, these kind of functions. So you will find actually they are very clever, right? So they can identify your voice and know what you want to do. So how these kinds of apps can do these kinds of things so accurately is because they also include the AI models in their apps. So that's kind of, you know, how the AI impacts this kind of daily life at this moment.

[00:04:45] Bouchie: So we have these applications in our daily life, but there are really also applications that are in the medical space, which we are going to talk about today. But one of the strengths of AI is that it allows us to process lots of data in a very short period of time. So can you talk about that a little bit?

[00:04:49] Yao: Yeah, so this is a great question. So actually, so if you look at the AI, actually right now, you know, it's a very popular and super hot topic. But actually, so, it's not that kind of new idea. So the AI concept that was initially introduced actually decades ago, but their application was hugely limited by the computational power at that moment.

[00:05:19] Bouchie: By the ability for the computers to do what they needed to do because of the technology?

[00:05:25] Yao: Technology was not there.

[00:05:26] Bouchie: Right.

[00:05:27] Yao: But now, actually, with the advancement of this kind of computer technology, basically, we focus on the two things, hardware and software. So they are so powerful now. And also, the other kind of component the AIs depend on the existing data set because they try to train their intelligence, the programs, based on this kind of data set. So now we're entering into this kind of digital era. So yes, now it becomes possible! So with the supercomputing, the computer is so powerful and also the data is everywhere. Why data is everywhere? Because we have the internet.

So the internet is a great platform and a vehicle to generate that kind of content in the data. And also, it's a great platform to distribute all these kinds of AI applications. So that's why it's time now that AI can really introduce into the daily life. And also, we have basically a lot of successful stories already. So one of the things that you will look at, for example, because the AI itself is a very broad discipline and you have multiple ways you can achieve the AI. So one of the ways you try to achieve that is - we call it machine learning. So machine learning basically is to try to build that intelligence program. And it's basically by allowing this kind of program to try to learn from the data set, existing data set, all the self-generated data set. And for a very complex problem, for example, you're trying to drive a car automatically or recognizing a very complex pattern. For example, just recognize somebody from the pictures, for example.

[00:07:24] Bouchie: Right, right. The facial recognition in our phones.

[00:07:28] Yao: So for those kind of applications, AI has achieved very promising results. Yeah, so that's why, for example, these kinds of things can apply to health care. For example, people already try to use this kind of machine learning approach, trying to analyze the CT or MR images.

[00:07:49] Bouchie: Right.

[00:07:50] Yao: Several groups already try to use this kind of approach, try to develop these kinds of diagnostic tools and to look at the pathology, for example, in the diabetes or the cardiovascular disease. Right. So... And for our lab, so yes, so we also try to use this machine learning, these kinds of powerful tools and combine with the traditional approach. For example, in our lab, we're using the multi-skilled biomechanics modeling, trying to study the musculoskeletal disease.

[00:08:25] Bouchie: Right, right. So the AI is only as good as the data set that goes into it, correct? So what you're saying is that because now we have this tremendous amount of data, now we are able to utilize AI in a way that we haven't been able to do before, because whether it's data around our shopping behaviors on Amazon, and Amazon can serve up to us something that we might be interested in shopping for that we didn't even know we wanted or needed, but also on the medical side, there's also a tremendous amount of data being generated there in terms of... of data around different health issues. You mentioned MRI and CT scan, you know, as far as x-ray imaging and those kind of things that allow us to utilize AI in the medical space.

[00:09:22] Yao: Exactly. So one thing actually, so we'll look at it here, is there any kind of, you know, data available and also if those data can be used. And so in the meantime, do we have the computational power to analyze those kind of, process those kind of data? So that's why at this moment, those two components looks like it's available now. So yeah, so the AI research or the real applications become exponentially with growth in the past couple of years. And we're envisioning actually for the next decade, coming decade, I think the AI could reach out to a lot of different kind of aspects of the daily life and also, we expect the AI and with this kind of powerful tools, they're going to have very beneficial change for the healthcare and research enterprise.

[00:10:22] Bouchie: Right, because we're going to hit a place where we have computers and technology reaching a level that they just simply haven't been at before. So in your lab, you're using AI, but you are a bioengineer, which we'll talk about that in a moment, but you are looking at TMJ. So can you just talk to us about what TMJ, or this temporomandibular joint is, and why it's important to you to be studying it?

[00:10:53] Yao: Yeah, so, you know, first actually, how we get into studying actually this very special joint. So... So the major reason actually is because of the faculty in the dental school. And so this, basically the joint actually is handled mostly by the dentist. So the TMJ, temporomandibular joint, and simply we call the TMJ, is a very unique joint. So that's the only joint actually with the one piece of bone, but you have two joints in the left and right. And also they are providing critical function related to our daily life. So for example, when you eat, when you speak, and even time, actually you want to laugh on something, and also facial expression, all have to use these two joints, the temporomedibular joints. Right. So unfortunately, actually, you know, a lot of people, actually, so they are trying to move their jaw not that easily. So... So for those people, actually, so they have so-called temporomendibular disorders.

So temporomandibular disorders are a group of musculoskeletal functional disorders and relate to the temporomandibular joint. And so the people with these kind of problems, so estimated in the United States, where about 10 million to 15 million people have this kind of problem. And also the TMJ also contributed to a large actually, basically a group of people with chronic pain disease. So that group actually is, you know, have a huge impact on the economy. So probably annual cost is around $500 to $600 billion annual cost.

[00:12:44] Bouchie: Right, because people experience a lot of pain when they have these disorders and often are seeking out their dentists in order to get support for trying to figure out how to get relief.

[00:12:59] Yao: Yes, because there are impacts on these kind of routine daily functions. So, right now the challenge is, yeah, so, yeah, it's a very significant clinical problem, but unfortunately, so, what’s caused this kind of problem? And in other words, the TMD etiology is not fully understood. To a certain extent, it's poorly understood. So that's why at this moment we don't have a very targeted treatment approach to handle those kind of problems.

[00:13:32] Bouchie: Right, so for a patient who has this issue, there may not be a lot of approaches for them in order to try to solve it or get relief because, to your point, it's not very well understood.

[00:13:46] Yao: Yeah, so the reason actually, we're not fully understanding the disease mechanism. And right now the treatment is mostly non-targeted. And so for example, we have the, you know, conservative treatment including, you know, with the pain medications.

[00:14:06] Bouchie: Right.

[00:14:07] Yao: And some kind of time actually in nerve block, you know, joint nerve block, and also a little bit kind of rehabilitation, you know, strategy. But it's only dealing with these kinds of symptoms, the pain, and also only offer the short-term symptom relief. And also from the surgical treatment side, yes, we do have the procedure called the orthognathic surgery or the other kind of craniofacial surgery can treat those patients. But the problem is a long-term outcome, is still very uncertain and many patients that are still gonna continue to have these kind of symptoms in the long run.

So, right now the need is to really develop this kind of targeted treatment and also have these kind of preventive kind of options, so we have to fully understand the disease mechanism. And currently in our lab we try to focus on the identified risk factors. And also to understand what kind of mechanistical relationships are between those risk factors and the TMJ, the joint mechanical functions. You know, by understanding those kind of fundamental relationships, we'll be able to develop targeted treatment strategies. So that's the kind of things, you know, we're doing. And the question here is how to identify those kind of risk factors. And also, how to understand these kind of risk factors, you know, go through what kind of pathways to impact joint function. And also, down the road, how can it impact the biology.

[00:15:49] Bouchie: Right. And I think that's a really great point. Shunchun, I want to turn to you for this one because I think you are a bioengineering student and you are looking at, particularly at the machine learning side of this and the biomechanics of it. And I think that's really important is when you're looking at risk factors, you're also looking at... you're looking at the biomechanical risk factors. What are the structures within the jaw joints that may put someone more at risk. Can you talk a little bit about why being a bioengineer in looking at this issue is helpful?

[00:16:32] Sun: Well, there are several aspects here. The first thing is that we are, as engineers, we are not only doing the engineering part. So as engineers, we typically seek a solution for a problem, and we are also doing part of the science problem. We are also trying to figure out what is going on in this world, what is going on within this patient. So to solve this problem, it is necessary.

So the first thing we are going to do is we talk with surgeons. We try to figure out what is going on in their observation. And we try to figure out what is the problem. So then we use engineering tools to try to target those problems. That is the engineering skills needed. And after that, we're going to use our engineering skills to really try to solve that problem. So although we are called engineers, it's really a combination of science and engineering skills together to solve this problem. And engineering can help us design instrumentations, for example, how we can track the motion, how we can efficiently measure the electromyography, those kind of key indicators, and it can also help us design solutions using our engineering skills.

[00:17:46] Bouchie: So that's really interesting because when we think about an engineer, we think about things like buildings, we think about bridges, we think about these type of construction... at least I do. But when you think about bioengineering, it really is a little bit of the same thing because it's about the strength of the thing and how the thing operates in space and in the world in order to be efficient, I guess, is kind of how I think of it. And so when we look at a mechanism like the jaw as it works along with the rest of the skull, you know, it's a little bit like the same questions you would ask about a bridge, I guess, maybe. Is it strong? Are there places where it's weak? How are those places that are weak?

increasing risk for things like pain or injury or you know these type of things. So can you tell us a little bit about what are some of the features of the of the jaw that are maybe predisposing people to have pain? Have you discovered some things that seem like pretty unique risk factors in the jaw that can kind of indicate this?

[00:19:00] Sun: Yes, of course. So I think I may want to start this with two interesting stories. So when we made the presentation at some conferences, as well as the scholars day at MUSC, we presented our work and someone came to us quite excited because either themselves or someone in their family has the feature we described and that is basically people with small mandible and people typically with the mandible shorter than the upper part, maxilla part, and typically, women are more likely to get that. And very often, people came to us saying, oh, I have someone in my family, or I am this type, and I do have TMJ problem. So that's exciting moment that we have.

So we identified, actually, what we find is multiple features. Just like I have already described three, there are even more to describe. For example, the condyle size is also another factor. It is exactly because there are so many factors, and each patient could have a unique combination that makes this teach thing so difficult to study. That is also why we rely on machine learning to give us an answer, because it's pretty good to look at a large volume of data, and to try to look at the very complex data to make a connection between things like a structure and the diagnosted result. So that is why we use this machine learning tool to study that. But briefly, the several factors that we identified, including the mandible size, people with small mandible are more disposed to... are more likely to get TMD.

[00:20:38] Bouchie: And the mandible is the actual jaw itself?

[00:20:40] Sun: Yeah, it's the jaw. The jaw itself, And also, women are more likely to get TMD. So that also matches the clinical observation results. And also, people with small condyle is also another risk factor. Those are just a structure aspect. We are talking about a structure. There are also other reasons. For example, people have been talking about mental stress as well as trauma. Those also could be problems as well.

[00:21:06] Bouchie: Because sometimes we might grit our teeth or we hold stress in our jaw, which then may lead to inflammation or other things that might exacerbate the issue.

[00:21:21] Sun: Yeah, that's also one. That is what we call, that is also one of the non-structural aspects is about behavior. Well, things like the one you said, whether we grind our teeth during sleep or whether we like to eat hard food, those kind of things. Those are related. So again, this is pretty multifactorial. That s why we start with the structural side. But actually, we're expanding the structural side to other sides as well to include things like behavior or stress, the other components into this.

[00:21:52] Bouchie: So that's really fantastic because that means as a bioengineer, you're not just looking at the strength of the structure, but particularly as a biomedical engineer, you're looking at the strength of the structure also surrounded by the behavioral forces that might be impacting. impacting that joint or in your case, you know, the jaw bone, but also or the jaw joint, but also, you could look at it in other skeletal features as well, I would guess.

[00:22:24] Sun: Yep. So the same technique could be applied to other joints, but because fundamentally the joints are mechanical system, you have structure to fulfill a function. If the structure has a problem... has a problem either originally or because of usage, you have problem in the structure, it will influence the function. So to recover the function, one of the best ways is to figure out the problem in structure and try to fix that structure. That is why we start with a structure and the structure is also related with how you use it. That's why, as you mentioned, the behavior is another important aspect of this thing. So structure and behavior and function, they are closely related concepts.

[00:23:10] Bouchie: So that's a really great point too that you just hit on and that I want to draw out, which is that the goal is really to look at how we treat what's happening at the source versus continuing with symptom management that might not get us very far. Dr. Yao, did you wanna say something about that?

[00:23:36] Yao: Yeah, so as Shuchun mentioned, so this is truly actually a multifactorial problem. So we try to understand actually how those kind of, you know, first identify the risk factors and also to understand how those kind of risk factors impact the joint mechanical function. As Shuchun mentioned, this, you know, the structure, behavior, and function, this axis.

So first actually, so we did this machine learning, and we identified several risk factors, you know, in terms of the structure. So the thing here is this kind of machine learning based work is great, it helps us to systematically go through all the data sets, identify these kind of morphological risk factors. For example, the mandible size, condyle size, and also ramus size. But beyond that, actually, we want to understand what's the mechanism, how it impacts the functions. So as a bioengineer, we have to understand and try to relate this kind of risk factor to the joint functions. So, one of the approaches we're doing right now, we try to integrate the machine learning and with the conventional so-called deterministic approach, the modeling. This is the focus on the biomechanics model. So combine the machine learning with the mechanics model to build the relationship between the risk factor and mechanical functions.

So that makes the machine learning even more powerful in these specific problems. And for example, so here we identify those risk factors from the machine learning, like mandible size. So we built a computational model to look at how this risk factor, for example, mandible size and the condyle size, could impact other mechanical functions. Is there an overload, generate a big joint force, or generate a big mechanical stress and to overload the joints? And also down the road, this kind of big force and big mechanical stress can impact the biology to initiate the tissue modeling and damage the tissues.

So those kind of things, actually, so we through so-called multi-skill biomechanical modeling to combine the morphology, mechanical function and the biology together. So you see this kind of multidisciplinary approach really benefits actually these kind of studies. And so here actually, so the point here is machine learning is very powerful. Without this kind of machine learning, these kind of powerful tools, we won't be able to systematically identify these kind of structure risk factors. But the machine learning itself also has some kind of limitations, because they are really not going to help you to build, to tell you why this is a risk factor, and what's the mechanical reasons and the biological reasons? So, by integrating with our multi-scale mechanical models to use the output from the machine learning. So we put it into the mechanics model and now we understand how it's going to impact the mechanical functions and how it will change the biology in the joint and eventually gonna lead to the disease pathways.

[00:27:16] Bouchie: Would it be accurate to say that the machine learning helps you see where to focus and then the other models that you have allow you to then investigate that focus where you may take years to figure out what the focus should be without the support of the machine learning processing these sheer quantities of data that it can process.

[00:27:41] Yao: Yeah, you just did the wonderful summary here. And so, exactly. So, without the machine learning, we wouldn't be able to search quickly to systematically identify the risk factor.

[00:27:54] Bouchie: Right, because the machine learning can't tell you this is the problem.

[00:27:57] Yao: Where it needs to be focused, yeah.

[00:27:58] Bouchie: It can say, here's some things based on your criteria that we think you should look at. This is what's rising to the top, so to speak.

[00:27:06] Yao: Exactly. So that actually really gives you the idea, where do you need to focus? And they're very powerful. They're not gonna miss anything.

[00:27:14] Bouchie: Right.

[00:27:15] Yao: So if we're based on the traditional approach, for example, based on this kind of hypothesis-driven and approach, conventional approach, we could miss some things. But machine learning is very systematic. It can identify all the risk factors in one time and overall. So then we have the focus areas. So in other words, in the engineering part, we have the region of the interest. Where do we need to be focused?

[00:28:43] Bouchie: The region of interest, yeah.

[00:28:45] Yao: So then we can study that better to understand the mechanism. Biomechanical mechanisms and mechanical biological mechanisms. So, then we can answer the question, what exactly the cause for this kind of disease or disorders.

[00:29:03] Bouchie: Right.

[00:29:04] Yao: So then we can identify the target, how to restore the TMJ functions using the therapeutic strategies, either through the medication or through the rehabilitation or the surgical treatment because we have the target now.

[00:29:18] Bouchie: Right, which is such an incredible thing for the patient because now we're making... we're able to make a more informed decision about what the treatment strategy should be because these targets now have been identified. versus maybe going down a list of what we think might work and kind of hoping that something would work. This is where we get to the more, the better possible outcomes by using machine learning and AI in patient care.

[00:29:51] Yao: Yes, and also the machine learning can take care of these kind of, you know, the complicated situations because each patient is different.

[00:29:59] Bouchie: Right.

[00:30:00] Yao: And as Shuchun already mentioned, so when we look at this kind of structure, you know, risk factor, there is all kind of possibility. For example, some patients only have one risk factor. Some people have two, some people have three, or it could be a combination of the different variations there. So machinery is so powerful. Yeah. So they learn things from the collective data set. But when they do the predictions, they can be patient specific. So they can identify for the specific patient, what's the combination, what's the problem. So it's very targeted.

[00:30:35] Bouchie: Right. So we've talked about the AI and the machine learning as a part of that. AI is the broad umbrella. Machine learning and also deep learning fall under that. But let's talk about the data that goes into that funnel. So when we talk about data that the machine is working with in order to develop these regions of interest where it says, hey, here's a flag in the ground, you should maybe take a look at this. What kind of data is going into that system?

[00:31:05] Sun: So for our machine learning model, we use a dental scan, the dental CT model. So that is our input. But that is not all. We need to process the CT in order for the model to use it, because AI is basically mathematical. So you can only put in things that could be represented by the mathematical entity. So what we actually did is from the CT image, we can get the geometry of the skull, of the bone. And then we try to discretize them into matrices. So now we have a matrix. The matrix could represent geometry. So if it is one in the matrix, it means there's a bone over there. If it's 0, there's no bone over there. Now we have a 3D, we have three-dimensional matrix, which represents the geometry. That is our input of the data.

[00:32:03] Bouchie: So you're taking an image, a picture, and you have to make it mathematical in order to enter it into the system to be used.

[00:32:12] Sun: Correct. So actually, anything that goes into a machine learning model needs to be mathematical. So even for other disciplines, maybe like computer vision, that's also something mathematical. It doesn't see a picture, it sees a matrix, basically a bunch of numbers. That's what machine learning could see, could understand.

[00:32:32] Bouchie: Wow, that's fantastic. So, you have the image, the mathematical image that you're putting it in. But you also are working with multi-dimensional data as well, right? So, what does that mean? What kind of data is that as well?

[00:32:47] Sun: Well, there are other things that could be used as input. So a very good example is the motion, because it's a joint. So, it has movable part to be a joint, right?

[00:32:57]Bouchie: It moves when it's a joint, yeah.

[00:32:58] Sun: Exactly. So that's why we put the motion data in, we basically use that as another source of input. And people are all very interested in the muscles, because muscle is the actuator for the system. And one of the good indicators for the muscle activity would be the electromyograph, which is the weak electricity generated by the muscle when it generates force. So that is another example.

And also clinical records and diagnosis result can also go into that, but maybe not always go into the input, but also go to the output as well. Basically, the machine learning model tries to build a connection between input and output. So, we can put that information in the input. Or we can say, well, it has this structure, and that we know from the clinical notes this is a female subject with TMD. So it could be encoded in the input or encoded in the output. So that is basically an example of other factors we are considering, we are trying to integrate into this model.

[00:34:08] Bouchie: And you're laying all of these data on top of each other. And rather than a team of people, two, three, five, however many people, sifting through all of this data, we now have the power of the computer to be able to plug it all in there and really come quickly to some ideas about where we should be exploring and gaining new research discovery around this.

[00:34:35] Sun: Correct. One of the problems with humans looking at the data, which has been there for quite some time, the problem is that sometimes we're not quite good at looking at something. For example, complex structures. We're not so good at looking at three structures and try to summarize what is going on over there. And it became even harder when we do it on the computer screen, because things were kind of distorted and scaled over there. But machine learning is different. It looks at mathematical entity. It looks at things... it could look at a big picture at the same time and it could look at multiple data at the same time. Think about someone, maybe I can look at two or three at the same time, but 100? No, never. But machine learning could do it. Could look at 100 very complex 3D geometry at the same time and try to find out the pattern and the connection with the diagnosis results.

[00:35:34] Yao: And especially when talking about the multi-dimensional data set, we mentioned that you have the CT imaging, and you have the motion data, and you also have this electromyograph data. And we put together how to look at it. And a human being probably doesn't have this kind of capacity, you know, at the same time try to look at all those kind of different domains, these kind of data, but the machine learning, yes. So they can look at this kind of the data set from the different domains and these kind of different dimensions. And also we look at, also the longitudinal data, not like one time spot. So we have different time spots. So they can dig into this kind of data set and try to find the unique patterns. So this is probably, the human being, probably, it's impossible to do that. So that's the kind of beauty to use in the machine learning. And specifically in this project and also this kind of approach can apply to the other kind of organ system and other kind of disease studies.

[00:36:43] Bouchie: Right, right. So, Dr. Yao, this project has been a tremendous collaboration across... it's a multidisciplinary collaboration. It's, we have the, we have dental medicine, we have a more traditional college of medicine, we also have physical therapy and some of the health professions. Can you talk a little bit about what it has been like to have this collaboration and what the strength of that collaboration has meant for the outcomes that you're seeing in your lab?

[00:37:13] Yao: Yeah, so as we discussed, actually, this is a truly multidisciplinary kind of approach studies. And so how to make these kind of things happen, actually? So here, actually, I would like to emphasize the one thing. So we have this kind of platform and we're able to group people with the different expertise. So in this case, we have the dentist. We have the medicine people. And so, we have the biologists to understand the joints, the biology. And of course, we have the bioengineering to be able to assess the joint functions. How are we going to put a group of these people and work together to address these kind of very complex problems? Yeah, we do have this kind of system to ensure the group people can work together.

So here, we have a very unique program. It's called Clemson University and Medical University Joint Bioengineering Program. So this program was actually established in 2003. And so the mission for this program is to try to encourage multidisciplinary translational research and education. So we are focused on the research and also we are focused on training our next generation of investigators. So within this group... the program, so we have a group of people, physician, biologist, and engineering, we are working together. And so in this case, engineers serve as a kind of liaison. And so they try to interact with the physicians. So when we start with TMJ, we start with the physician, and we try to get the first input to define what's the clinical problems. And in the meantime, we reach out to the biologist and to see these kind of mechanical functions and joint environment – how do they relate to the different biology down the road. And as a bioengineer, we can put all those kind of output from the physicians, from the biologists, and put it into our computational model and simulate the whole joint systems, identify the risk factor, and also to elucidate the mechanisms. Based on that understanding and with this kind of knowledge, now we go back to work with the physician again, so try to design new strategy for the treatment or for the prevention.

So that's the kind of unique platform through the Clemson-MUSC joint bioengineering program to ensure we have the expertise, multidisciplinary expertise, and also the workforce and investigator, as well as the students in the program to work on this multidisciplinary project. So for example, we do have expertise for the machine learning from engineering part, but we also have the clinical input to help us to identify all the... to classify the data set with the diagnostic outcomes. And also, we work with the investigator from the College of Medicine on a standard biology. And also, we work with the health professionals and look at the different strategies for the rehabilitation through the physical therapies. So this is such a unique environment. We are the only, probably, the program with the engineering, bioengineering on the medical campus and work with the team like that. So it's a very unique program.

[00:41:03] Bouchie: Yeah. And I think your point about that it really is research that moves towards translational is a great one because what you're doing in terms of learning about the jaw and the risk factors, a lot of that is through the process of the data, you're discovering things that we don't know yet about these joint systems. But it's towards an end of moving towards... moving it to patient care and using that information to improve how we care for folks who are experiencing disorders in the jaw. And that's so important that we can make that connection and show that this is what research does, is it discovers the things that we don't know so they can move towards improving care and prevention in the future.

[00:41:55] Yao: Right, so as an engineer, we always try to translate all the discoveries, all the findings from the lab and to the bedside. So our ultimate goal is basically to treat the patients. Or we could develop a strategy to prevent this kind of disease. So we believe as an engineer, because as Shuchun mentioned, so the engineer is, the essence is to try to focus on the problem-solving. But problem-solving, we also need to understand the mechanisms. So these kind of problem-solving strategies are based on well understanding of the disease mechanism. So that's why all the strategy we develop is very targeted and available with very sound science as a backup.

[00:42:46] Bouchie: When we think about AI in this space or just AI generally really, I think you have some really great examples to talk about this as a tool. It is not a tool that takes precedent over anything else, but it is a tool that we can use to expand our learning and discovery, which of course is something that we really love to focus on here on Science Never Sleeps. So Dr. Yao, can you talk a little bit about AI as a tool?

[00:43:15] Yao: Yeah, sure. So, AI actually, you know, just like, you know, other tools, you know, so human beings actually invented and over the years have tried to understand better about this world and tried to create the tools to develop the things that really benefit this society. So look at the artificial intelligence. So compared to the human intelligence, they do have a lot of advantages. For example, So the AI can really learn from very complex patterns, and they never get tired. If people look at it , they get tired and start to make mistakes. Yeah. And also, if you look at actually... other things over the years, actually, human beings always try to create these kind of tools to... to benefit actually and make ourselves more powerful to observe the world. So, to give an example, human beings, their vision capacity is limited, so they cannot see very, very far away the things. So that's why they tried to create some kind of the tools. Eventually, they invented the telescope, so they can see the stars and see the details. So just recently, you know, so we put a very fancy microscope in orbit so we can observe the far away galaxies.

[00:44:56] Bouchie: Right.

[00:4457] Yao: That's just the tools, but that tremendously increased actually the capacity for the human being to observe things far away and the same thing actually, so we also try to look at the things in the very detailed, very small scale, but our eyes won't be able to do that very well. So that's why we... come out the microscope. So it's a very exciting thing. Now we can see the little microorganism, like a microbial, and those are very clear details. So that kind of image helps in the biomedical research tremendously. The same thing here, and now we face new challenges. So, we have such, kind of, fruitful data set and very complex patterns to identify. It could come from this CTMR imaging, could come from the motion, and come from the electromyograph. Or it could be a combination. And how are we going to deal with that? And human beings probably, with their bare eyes, probably is hard. We need new tools. So in this case, yes, the artificial intelligence specifically could be the machine learning tools is a geat tool, just like a telescope or the microscope, help us to better process those kind of data set to get meaningful results. And so there's no surprise, I think, down the road, that the human being will create more tools to help us to better understand the world we're in at this moment.

[00:46:39] Bouchie: I love that. I think it's a little bit like the... AI is the microscope for the data scientist.

[00:46:47] Yao: Yes, of course. So we can understand... you know, consider actually, AI like that.

[00:46:52] Bouchie: Mm-hmm.

[00:46:53] Yao: Yeah, it's a powerful tool. And also, we can ensure this powerful tool can be used properly.

[00:46:01] Bouchie: Mm-hmm.

[00:46:02] Yao: So that's why we have all those kinds of different strategies to guarantee those kind of tools in people's hands with a good cause.

[00:47:13] Bouchie: So it feels a little silly to ask what is the future of AI, because it feels like the future is AI, but certainly there is a future to utilizing this tool. So what is the future of this in biomedicine?

[00:47:30] Yao: So first, actually, for sure, AI is a very powerful tool, and also can make very significant change the way actually where... doing the biomedical research and also the health care practice. So no doubt. And we envision actually these kind of applications become more and more with these kind of exponential growths. So one thing actually I would emphasize on that is we need to notice actually first how powerful these tools are, and also need to understand. any limitations in these two. So that could help us to fully utilize and explore those kind of the potential for these powerful tools. So one thing actually I'm thinking about in the biomedical research, so the AI must be combined with the traditional approach we are using at this moment. For example. how to combine with the way of doing things, you know, daily based with the, you know, using the, you know, batch top, you know, cell models, and also with the animal model studies and the clinical trials. So in that case, we can ensure this kind of, you know, outcome from the AIs and it's meaningful, and with the great magnetistic understandings. So this kind of output, when we go back to use for the patients, so we have fully understand the mechanism. So it's basically, so we're ensure this kind of outcome so we use this properly.

So the second part, actually, so we're looking at this and how to increase this kind of application for the AI. As Shushan mentioned, so the AI is based on the data. So how are we going to create this kind of mechanism to better generate this data and share this data? Yes, we're in the digital era, right? So we have a lot of the platform being able to generate the data and share the data. And also especially try to create the multi-dimensional data set. And the one thing I would emphasize and also be cautious is when we try to collect this data, and store this data and use this data, we have to make sure we're doing this in a very safe way. Because the tool is powerful, but we need to make sure the tool is in the good people's hand and used properly. So that is what I'm thinking about. But overall, I envision the AI. And it will probably lead to significant change in biomedical research in the near future.

[00:50:35] Sun: Yep, so totally with what Dr. J.L. mentioned. So I think in the future, the thing that we really want to see is really large quantity of data and more diversity data. So we see all kinds of data could be used as input. As the biomedical engineering develops, we can gather more data. We can have more information about the patient. Those information can all go into the machine-naming model and contribute to the health care.

[00:51:05] Bouchie: Thank you so much for joining us on Science Never Sleeps.

[00:51:09] Yao: It's great to have you here. Thank you for having us.

[00:51:11] Bouchie: We've been talking to Dr. Hai Yao and Shunshun Sun about TMJ and the use of AI to improve treatment and prevention. Have an idea for a future episode of Science Never Sleeps? Click on the link in the show notes to share with us. Science Never Sleeps is produced by the Office of the Vice President for Research at the Medical University of South Carolina. Special thanks to the Office of Instructional Technology for support on this episode.