BoneChat #96 - Implant Identifier app on your phone to better plan revision hardware removal.

So this is Bone Chat 96. We've got Dr. Parth Desai. Is that right?

Um, perfect.

Pretty close. Uh, he's a new spine fellow, just graduated from Northwestern. He will be, uh, a spine surgeon in Atlanta, was first, first real job at Piedmont, right in Mark's backyard. And he's super entrepreneurial. He loves AI. He's wearing the Meta glasses. I mean, we're going to go through his implant identifier, which is— why didn't we all think of that? And then we can talk about augmented reality and AI if we want. So I'm turning it over to Parth.

Absolutely. Thank you so much, Tiger, and thank you for having me.

Yeah.

So the, the problem sort of is one that, that any surgeon, any rep that's been in our space for any amount of time has come across. When you're planning for revision surgery and you have prior implants in the patient, you want to know what exactly those implants are for a number of reasons. Not the least of which is having the compatible implants available, having the removal implants or removal equipment available, but also just more nuanced reasons. Why did these implants fail? What overall nationally are the revision rates of different, different implants that we see? And just so many different things that, that this information sort of tells us. Um, so about 8 years ago, uh, when I first, uh, when I was actually a medical student at this time, when we first came across this problem, um, I remember reading papers about, uh, the idea, the concept of using artificial intelligence to identify implants just based on the X-ray using sort of the morphology of the, uh, of the implant and, and doing that sort of in a way that would be instant as opposed to sort of the current method right now, as, as I'm sure surgeons and reps that are listening know, it's a bit of a process to find what these implants sometimes are. In the best case of scenarios, if we have the prior op note, we can look at the op note and then usually get a sense of what the implant is if it was in there. In the scenarios where we don't have the op note, if the patient's coming from another town, another country, a different EMR system, that's when it becomes a bit of a challenge. So we're texting out these X-rays to all the reps that we know. We're asking colleagues that we have. There's a couple of different social media sites, Facebook and Instagram, that reps and surgeons can send these X-ray images to and hopefully get a response. There's, there's old books that, that have some of these implant X-rays that if you, if you have the time, you can flip through these books and see if you can find a match. And really, you're hoping that by the time your surgery comes up that, that you figure out what it is. So that's—

Yes, it's like tribal knowledge. That's really what it is. You get, you find an old orthopedic surgeon that's been around for decades and he or she can maybe identify it.

A lot of times it's crowdsourcing. Exactly, exactly. That's the best thing that we can do. And, and there are some, there's some reps and surgeons alike that I'm very impressed with, with how vast their knowledge of these old implants are. But again, that's, that's few and far between often. And especially if you're a junior attending, a junior, a junior rep, a resident, you probably haven't seen enough implants to know really what you're looking at. So it can be a problem and it definitely is. So about 8 years ago, we first sort of started working on this concept of how do we build a model that can identify sort of any implant. Our focus initially has been on hips and then secondarily moving to spine, knees, shoulders, primarily arthroplasty, but then we're also wanting to break into the trauma side. So the challenge just with training AI is you need a large quantity of high-quality data. So you need labeled implant X-rays of each kind that oftentimes, depending on how similar these implants look to each other, which in the hip space, just with the FDA 510, we'll see that a lot of different implants within the same class from different companies look very similar with often just sort of very subtle differences in their, in this or that. So we needed a large amount of this data. So initially myself and some of the others in this space that I've been working with worked on getting sort of partnerships with hospital systems that are doing high volume of certain implants and used that to, to sort of collect our initial datasets. But what we found is that there's just too many implants and the implants that we really needed to build out this model are more of the legacy implants that surgeons aren't putting in on a daily— on a regular basis now. They're just coming in when these need to be revised, maybe 20 years later. So the best thought that we had for how to solve this problem was, what if we could crowdsource the entire issue? So allow surgeons to utilize an initial sort of preliminary model that was rudimentary, just sort of gave maybe a large number of possibilities for a given implant, but did narrow it down some, and then allowed the user to sort of tell us what the implant ultimately was. And this we released about 8 years ago. So over this time, we've collected hundreds of thousands of images. Of those, a portion of them are high-quality labeled images that we were then able to feed back and create these models. The biggest thing that I found that we were able to put together is a really large number of like-minded sort surgeons and reps that are dealing with this problem and bring them all to one platform. So as of now, we have over 27,000 surgeons using the app, around 15,000 device reps, and then some number of just sort of other healthcare providers in this space or others that are in this space but may not fit in the larger two categories. And with this community, the idea sort of came, we have these Instagram groups and we have these Facebook groups that have been really valuable for many years now in terms of crowdsourcing the answers. And why don't we just put all of that under the same platform as we have the library, as we have the filters, and as we have these AI models. So, so that's been the new update of Implant Identifier, which actually just released a couple of weeks ago. We call this sort of a soft release. We're getting valuable feedback from our users and making updates as we go, as we kind of prep for our full release to our over 50,000 users at this point. So with the current update, which I would like to share with everyone today, we have a fully functional HIP detection model that can identify any of the most common 150 revised HIPs hips with a high degree of accuracy. Though in certain cases, we're just sort of narrowing it down still. But again, as we narrow these implants down, as the user is able to confirm with us what the implant actually is, this feeds back into the model and continues to make it more accurate. I'd like to give a demonstration of that today with just some hip implants with— we have which we have.

Yeah, real quick, Emanuel's got his hand up.

Yeah, just, uh, very quickly, can you guys hear me okay? Yes. Okay, cool. Sorry, I had some problems with my mic earlier. Um, are you, um, are you in touch with the guys at Yale, with Lee Rubin and those guys?

Uh, I've not worked with them. Uh, they're also, I take it, working on something similar?

Uh, no, not something similar, just it's a, it's a real interest of Lee's. He's gathering, uh, older implants so you can kind of— so there's a library. They do about 2,000 revisions every year, and they have to store those revisions for 7 years. So they've got about 7,000— sorry, about 14,000 explanted units in the bowels of some storage cupboard somewhere in Yale. It just immediately made me think that could be a treasure trove of older implants if you were looking for, you know, more data input. So I just wanted to mention that while it came to mind.

Yeah, no, that's a great point. Explanted implants are just as useful. You can take an X-ray of an explant and it looks the same for our purposes as far as identification goes.

Once the call's finished, if you want an intro, I'm happy to kind of put you guys in touch and see if there's an opportunity for anything to be shared there.

Yeah, yeah, that'd be wonderful. That'd be wonderful. Greatly appreciated.

Eminol, I put Parth's email in the chat.

Yeah, I'll take it. Okay, cool.

Thanks.

Take it over, Bart. Very cool. Yeah, so, so yeah, any, any other questions, please ask while we're, while we're going through the demo. Just, just go ahead and go right ahead. I'm going to go ahead and share my screen here.

Steven's got a question.

Sure.

This is awesome.

How long has this been live?

One.

And I think this is just going to be a huge win for so many reps. I've seen so many cases where as soon as that R revision word comes up and you got to take something out, you just sit there on your phone trying to look at or find X-rays. And this is, this is a great, a great win. Yeah. So between just our sort of internal testing group, it— we released it about a week ago and, and have been debugging. Our full release is still sort of rolling out, but when I could say the full release, that's when we kind of reengage all of our current users using the old version and transition them over to the new version. Just the main thing being just appropriately debugging it at that point. Yeah. Well, I love it.

Yes. After—

I think there's a fun opportunity for something we're both doing here with the same, the same users. Yeah, yeah, no, absolutely. I love it. Um, so, so here we have, uh, so I'm in the app and, uh, we're looking right now at our community feed. This is where our users can, uh, can post a, uh, an x-ray of an implant they, uh, they don't know and get real-time feedback from, from experts in the community. Michael Cassioto being, being someone that I know well and know that is very good at identifying implants and others as well in our community that are quite good. So this is just sort of the new feature that we've kind of rolled out, just allowing, as I said, users to post implants and get an answer on what it is. Moving on to our library. This is another sort of new addition to the, to the old application, is these pretty advanced filters that we have. So these now we can, if you do have characteristic information and know the shape, then you can actually filter out to narrow down your search if you're using sort of the manual identifier. Forgive me, my internet seems to be a little slow right now.

Parth is actually in Munich right now, so he's 9 hours ahead. It's that lousy Germany Wi-Fi.

It seems to be, seems to be working okay though. So yeah, so as we As we look here, we have, we have some different things that we can do. We can filter to, for, uh, to narrow our search if we're manually looking at the implants. One thing we have in a number of these different implants is we have the removal information, specifically on spine. This is something that users can also, as they, uh, are using the app, can add information that they want added to the application. So once they, once they're going through the library, if there's something that they want to add, like this is a useful tip that, that I found when I was revising this case, they can add that and that's there for, for, for future users. We have a, we have a My Implants section, which was something that, that we've added with this new update, which saves any implants that you previously previously submitted to the app, whether unidentified, identified, or posted on the community. And you can access these at sort of any time. Now, here's an example of— this was a real implant that was submitted that immediately was able to tell what the implant was despite there being a calcar fracture through the stem. Hmm. And so here's some sort of some examples of, of how the predictions might look. Once you submit the implant, it's going to give what it finds to be the closest similar implants. In this case, the first suggestion was correct, and then the user can, can click on the implant card and, and confirm it. And then this sort of feeds back into our system. So now I've confirmed it.. It's going to now show up on, on the user's identified tab. And on my end, it's going to, it's going to show up in our training set. So, and I do want to, I do want to show some live examples just for some random implants from Google. But just to give a sense of what the backend looks like when, when we're submitted on the app, this immediately goes to our admin panel where our team can validate these implants and make sure that the AI is getting good quality data, which is really core to any AI project. What we all say, garbage in, garbage out. If you're not getting high quality data that's being validated, then your models aren't gonna be accurate. So, so this is just again our admin panel, but everything feeds back into, into this. And so we look at the implant that I just submitted. It's here. And then if I can confirm it, I just click yes. And then that feeds into our training set. We also have— we also allow our users the ability to submit new implants that have not been— that are not currently in the library. So these So as these are submitted, I see them here. We typically have a validation team that just validates that these are the actual accurate implant. But as soon as they're validated, we simply click Add Implant and this feeds right back into the library. And now this new implant is added in the library. So over time and just with the good graces of our community of users, we're going to be able to continue to grow, continue to find even some of those really rare legacy implants that aren't that commonly seen and feed those back into our library. And that's one of the pretty exciting things here. So just to give an example, we're just going to look up hip implant X-rays on Google and see how we do here.

So anybody want to guess before the identifier?

Yeah, any ideas? Reclaim. You thought one of them was reclaim? Okay, let's see. That may pose a challenge. So what I'll do here, so we're gonna, we're gonna test these out. So we think this is a Reclaim and this is another one. So let's see, I'm just going to take a picture of both of these. 1 and 2. So we have 2 options when we when we're submitting an implant to the application. You can use your camera to take a picture live, or you can, you can pull an image from your camera roll. So in this scenario, we're going to just pull it from photos. These are the last two. So we're going to first try this implant here. We zoom in, try to, try to get it centered. It doesn't have to be perfect, but, uh, this also helps when we feed the implant back into the, uh, into the model. So as you see, we have 3 options for what to do with this implant. We can use the AI to identify the implant, we can go straight to posting it on the community feed, or if we know what the implant is, we know it's not in our library, we can submit it to the library. So once we click AI identify. Our AI is going to analyze our, our implant and give us the possible suggestions. In this case, it is correct. This is Anthology high offset, and it gives a couple of other examples. It also gives you the option of, of again later posting it to the community or to the library if you're not able to locate it. We confirm it and And we're done. Same thing with the— that other implant. So we'll do the same thing here. Try to zoom in a little bit. Good.

It's really fast.

Yeah, so this one, um, I'm gonna have to trust it. I believe this is gonna be an Arcos with calcaneal. Exactly, sort of continue here.

Yeah, what's impressive about that is there's 5, there's 5, 10, 30 versions of every stem because everybody copies each other, but it's cap— what's capturing that you don't necessarily know what it's capturing, but it's finding some sort of nuance or feature that is particular to Archos or to Anthology. Does it ever tell you, or is there a way for you to find out, like, what those flags were or what the key points were? Because there's also a potential that— and maybe this isn't built into your system yet— but there's also the, the mating components element that can come into play, where if you know that every version of an ML taper type of stem doesn't take a big ceramic head, but only the Anthology one does, you could look at the other parts in the X-ray and help identify what the stem is.

Exactly. And that's something that we're kind of working on, on getting that data. And that's where I think getting some buy-in from the, from the companies themselves and from the reps that are using it to help, help give us some of that information. One of the challenges is that just with With most AI projects, there is the, what we call the black box problem where we don't actually know what it is that the AI is looking at when it's, we can, we can make conjectures. It's typically the same things that a human might be looking at when they're trying to, when they're trying to decipher the differences between two stems. They may be looking at the calcar, they may be looking at any holes or removal sections, and then certainly the distal stem itself. Are some of the key components. But yeah, no, it is, it is, so far you've got, yeah, so far you've only done AP shots.

It'll, I assume it'll work laterally too. But hips are probably easier to identify AP. And same thing with knees.

Do you do AP? Great, great question. For hips, we just utilize the AP itself. For shoulders, likewise, just the AP itself. For spine and for knees, we do use the AP and the lateral. So first, we take a picture of the AP and a lateral. And we actually have two models that then sort of narrow it down based on both of these results. So yeah, no, great, great point, though. But luckily for the hips, we typically don't need—

The lateral, yeah.

Don't need to rely on both. Exactly.

That's the amazing thing to me is there's no dimensions. So you would think the AI would say, okay, what's the size of the acetabulum, you know, cup? And then they would backtrack, but it just doesn't need to know dimensions, which is incredible. It just knows shapes.

Yeah, yeah, exactly right. And that is something that would be helpful, being able to know the taper sizes and, and well, the tapers on these implants. But then, and I do think that that's something that is coming and something that we can expect to see in the future. The one thing that that would then have to include is some sort of calibration marker in the X-ray.

Yep. So I'm dying to—

I got to ask. Is that I think is of value. So these, again, that's the, uh, our typical, um, yeah.

Cool. I gotta ask two questions. Uh, how are, how are you guys funding this and how big is your team? And then the second thing I'm trying to ask is what's the business model? How do you guys make money?

Yeah, so this is really more of a passion project and so entirely self-funded. A little bit about my background before I started medical school, I had, in between undergraduate, I started an app development company and a healthcare IT consulting firm. So through that, we built a pretty good team for developing these sort of products. Through medical school, I kept my my team and started focusing more on this project and some of our own projects. So yeah, so our team right now, we have 3 full-time engineers. Most of our team is, like most tech companies, are based overseas just for logistical reasons and for cost-saving purposes. But we have 3 full-time engineers, 1 full-time designer, We have a web development team that works more on a, on a per-needs basis, and we have 2 data curators, and I'm working on actually currently hiring a social media marketing consultant, so someone that can just manage all of our social medias. So yeah, again, this, this isn't, this isn't my day job, so this is something that we really just want to see grow and see where we can take it just in terms of, in terms of how big of a community we can build. I do think long term, just more in terms of recouping our investment, there are going to be some avenues for that. Some of the insights that we're able to gather for that can be helpful for the big companies is if surgeons opt in we can actually connect surgeons that are revising a specific implant to the sales reps and device reps in their region that represent that product, just to make sure that they have a direct line of communication with, with whatever reps they need to revise this. So these, these sort of insights, I think, are pretty valuable for a company to have if they know, okay, Dr. So-and-so is revising one of our stems on Next week, let's just talk to him. Let's be a little proactive and make sure that he knows that we have everything ready for him and we're fully, fully prepared. And these sort of things I do just as a surgeon, I think they do make a big difference when you're deciding what implants to use, specifically on the spine side. Oftentimes if I'm revising something from another company and that rep especially, I mean, for me, a pedicle screw is a pedicle screw, a rod's a rod. I do have certain companies that I use fairly exclusively, but for revisions, if there's an implant that's already in there and the rep has done a good job of showing me that they're going to be ready, I might just use that implant just to, just to sort of make it easier. And that is something that I think is a valuable insight that the companies are going to want. From the identification standpoint itself, once these models are really sort of optimized, I do think that's something that people would pay for. Whether or not we actually charge for it is another story. But yeah, it's— what I found is that by keeping it free, this is something that the Instagram, the Facebook, all of these insights, even just the text threads of, hey, anyone know what this is? This has all been a real community effort. This has been something that we do just to help others and to help eventually help the patients that are, that are having these implants revised. So I think just keeping it on that note, keeping it free and keeping it for the community has been helpful in bolstering growth. And I want to keep it that way.

Got it. Yeah. Maloney's got his hand up.

Absolutely, please.

Yeah, let me go off mute here. So, um, you grew up in Atlanta, Georgia, it looks like. Went to University of Georgia, I see. Mhm.

Yeah, go Dawgs!

You hear me, Barth?

I can.

Can you hear me? Yeah, I can hear you.

Oh yeah, okay, good. Yeah, I do. Okay, so I got 3 questions. When did you, when did you start using AI with this? That's one question. Number 2 is, actually I got 2 questions. The monetization question's already been answered. The second question is, what are you doing as far as desanitizing, de-identifying, or I should say de-identifying the images down to the meta level. And I'm curious about that.

Yeah, no, great question. I mean, I'm glad you mentioned that. Privacy is obviously a big deal, both to protect our patients, to protect our users. The primary thing that we do, and this is something that's kind of limited us right now to just being a mobile platform, but on a user's mobile device prior to them uploading any image, we have a proprietary text removal model that works locally. So once you download the app, it's on your phone and it runs every image through this model. So any amount of text that's on the image is removed. And if we find that there's too much text on the image, because again, these are X-rays, occasionally they do have— someone might accidentally have words somewhere in the corner of the X-ray, but typically we're not, we're not wanting to see something that's just text. So the, the model will actually remove that. But yeah, we have a model that works locally to remove any, anything that's deemed as PHI before this gets uploaded to our servers.

So if you're, if you're curious, I have a partner at Salt Lake that can de-identify all the way to the meta level. And they're, they're using this at Emory and some other places. It's goes far beyond just text removal, should that ever be of interest.

Very interesting. Yeah, no, that, that can come up. Right now, since we are relying on the mobile device as sort of a transition point between taking data from PACS to, to our servers, luckily we're, we're able to avoid some of those Gotcha. Some of those issues. Yeah. That's, that's great.

As Tiger knows, I have a, I got a long history in this that I was one of the founders of one of the first ortho digital imaging companies.

Is that right?

So yeah, I'll shoot you an email on it just so I can introduce you. Yeah. What about when, when did you, when did you add AI to this? 'Cause that's, that's the sauce.

Yeah. Yeah. So, so our, our first AI model was in, was in 2019. And this was a preliminary hip model. It basically just took any of the 20 most common hip implants and parsed those into categories. As we used this data that came from this initial model, we were able to really make highly accurate models to the one that we have now. And this hip model has been available just now for a couple of weeks. Our next steps outside of perfecting this hip model are to move into the spine space first, and then knee and shoulder. We already have actually preliminary models that are still similar to that initial hip model, going to be used to just collect data and allow us to feed that back in and create even better models moving forward.

Right. Okay. And last question, or not actually, just comment. If you need a social media guy, uh, I got a guy in Kyiv, Ukraine, who is probably one of the top LinkedIn experts in the world. And because they're in Kyiv, they're obviously pretty affordable.

Yeah. Um, yeah, let's definitely talk after this. That's, that's exactly what I'm, I'm looking for, someone to kind of, um, help take some of that stuff over, um, particularly as I'm— as I transition into starting my practice, uh, a lot of this is gonna hopefully— and I, and I have a great team that, uh, that really sees the mission and understands the spirit of what we're doing. So I can, I can sort of let them, let them sort of—

that's great. Well, you got a great story. It's— I really enjoyed it.

Oh, well, thank you, thank you. I do appreciate it. And, and at the end of the day, I think, um, what When we talk about one of your questions related to monetization, I think at the end of the day, this is more of a data company than just an identification company. And if we did, and one of the common models for something like this would be like a freemium model where we have some resources that are always going to be free, the community feed, the library itself. And then have a second tier of resources that cost a monthly fee. Say the AI models may cost some nominal monthly fee. But really, the thing that we have that I think is most valuable is the data that we're getting in. And again, this is, as we grow, we're getting implant data worldwide of what implants are being revised, what are the modes of failure. And this I think is, valuable in ways that we sometimes don't fully understand yet. As we see different insights that we can get from this data, I think then we're really going to understand the next generation of products that can come from this.

So that's—

Yeah, you're almost—

That's why we should talk at a different time because big data is a new oil field. And that, that partner of mine that can de— that can sanitize your images is critical to a really usable data lake.

Absolutely. Absolutely. And yeah, the sanitization is one core part of it. And then just structuring the data and organizing it on the back end, because that's, that's a problem that I think now we're becoming more cognizant of. We had for so many years so much data that we really that was just unorganized. It didn't really have necessarily a purpose or a structure to it. And we've seen that in all areas of AI. And now the focus really is on not just having data, but having structured data that we can actually do something with, which is really, I think, was why this new update and this new version of our app was so important to me that For years we were collecting all this data and we were getting some pretty good labeling of it just in terms of what these implants were. But I just knew that there was so much more that we could be doing with the data that we're getting in. And that, that really is, is part of why we've kind of built, built out this community forum section where as users have conversations about these implants, we can take that and make that into meaningful metadata associated with the X-rays.

Cool. All right. Emanol's up. Oh, you're on mute.

Okay.

Sorry about that.

Yeah, I guess a couple of questions. One is anybody kind of submitting info that you're then going to use when you aggregate the data? Are they signing a release or They're kind of checking the box saying they're okay for the data to be aggregated and used.

Exactly right. And it's a pretty— our terms are pretty— let me one second. I'm going to share this with you. These are the terms that have to be agreed to with any image that's updated. This has to be manually checked. First of all, it's just a certification that there's no PHI. But when we get into the terms, I mean, we're pretty liberal with what we're considering data that's submitted to us. If you look at it, we're granting— by submitting an X-ray, we're being granted non-exclusive, royalty-free, perpetual, irrevocable worldwide license to use, reproduce, display, and modify. This essentially is characterized as the most— essentially the maximum utilization of these, of the images that are submitted to us. And I mean, it's a bit of a gray area right now in terms of what this data means. The question, I mean, the question really hasn't been— the precedence hasn't been set for who owns the data right now. The best that we can understand is that the hospital system would be the purveyor of this data, with the exception of for research and educational purposes. But it's not the patient that owns their X-rays. They can use them, but it's not the X-ray tech that took it. It's not the X-ray machine owner. What right now— the best guidance that we have from this nationally is that the hospital system owns this data to some degree, but the surgeon can use it for educational purposes, for research purposes. Is de-identified, which is sort of what we're—

I raise it because my, my company's in a similar situation. We do bioengineering analysis of explanted prosthesis, and then question of, well, who owns the data once that analysis is done? Because we have all the data on, you know, taper failures, wear junctions, all kind of all the rest of it. So that's, that's a minefield we have to kind of step through as well.

Absolutely. Um, and, and I'd love to hear more, uh, later about how you guys are navigating that Yeah, from our end, I'm hoping at some point we start seeing some guidances related to this. And I'm hoping that the guidances are more supportive of the research that's being done around it, particularly when we talk about just the AI race and places like China where essentially all the data can be aggregated and used by these companies. We have to, build an environment and an ecosystem that's beneficial for the researchers that are trying to help the patients. So yeah, and that's, I think, things that we're going to see in these next couple of years.

And then my second question was around monetization, because I understand the desire to kind of make this kind of free at the point of service, if you like. We we support the NHS in the UK and we try and be as cognizant as possible of the NHS budgets and kind of trying to provide a service to those guys. But I wondered with you guys, is there a model that allows you to charge a nominal one-off kind of joining fee, literally just to kind of COVID your costs? Because, you know, it's not— nothing's happening for free. And if you do want to avoid kind of freemium models in the future, and, you know, if the app needs support, at least you've got the opportunity of running it as almost like a B Corp, you know, or a not-for-profit where you are at least washing your face in terms of making sure the app is, you know, kind of supported enough for the, for the growth that you aspire to.

Absolutely, uh, no, absolutely right. And, and at some point that may be where we're, uh, where we're headed, having to charge some sort of monthly fee for, um, for access to the models.

Right now I wasn't even thinking a monthly fee. I know there's a couple of apps that I use from a, like, an exercise perspective. I think I paid a one-off fee for access to the app, and then that was it, done. But, you know, it was— it's a nominal amount, but at least it supports the development of the app.

Just a one-time sort of lifetime, um, fee. And, and again, that, that is an option too. Um, it's, uh, it's really just a matter of what, um, is how many users are we going to keep from joining the app by even, even say a $5 or $10 fee? Is that going to keep people from joining the app? Is that then going to keep us—

people pay $7 for a latte.

Yeah, yeah, yeah. So no, I, I think, um, I think people would pay, especially— I mean, our users are surgeons and device reps. We're not, um, we're not just, uh are marketing to the masses. So this is a very niche user base that does have the budget to pay for this. So yeah, no, certainly that's something that's come up right now, especially as we're still working out these new models for spine, hip, or spine, shoulder, and knee. We definitely are just at this point going to keep everything free, get as many users as we can. Ideally, at some point, if we can get to 100,000 users, That would be wonderful. And then we can start thinking, okay, some percentage of these users are going to have to pay something because at that point then our server costs start becoming pretty, pretty—

Well, at that point you've also got great data as to, you know, what percentage of folks are using it regularly and who's likely to kind of notice the absence if they don't have access anymore.

Exactly. Exactly right. And that's, and that's the other thing, just charging that and then still having the models charging some nominal fee where if the other aspects of the application were free, but the model was maybe $5 to use one time or $7, say, to use for the whole month unlimited, I think a lot of people would just say, you know what, even if I'm just going to use this once, I'll just go ahead and pay the monthly. So there's definitely options. I think that this is something that there's enough people that struggle with this issue that people would pay to have their implants identified accurately.

It's such a cool solution. I can't see why folks wouldn't.

Well, thank you so much. Yeah, yeah, no, that's what we're hoping.

But— So Andrea is up and then Kyle.

Yeah, you were talking about China and the AI use there.

Is it—

is this app currently mostly focused on US implant companies or is it international? Yeah, so very much international. I would say our highest percentage of users are in the US and Europe with close following of South America. And Asia markets. And then again, we do have some, some African users as well and some users in the Middle East as well. But, but yeah, it's really intended to be a worldwide application. Our big 5 companies are pretty widely in use with the exception sometimes India and China have their own companies that kind of make knockoff of some of the big company implants. But But yeah, it's really something. I've been sort of surprised at times where we sometimes get new users joining from like just company or countries that you may not think that they're doing a lot of joint replacements. And then you see at their government hospital, maybe someone's revising something and they utilize the app.

So yeah, I didn't mean so much users as the implant companies, right?

So if there are these knockoff companies in India and China wherever, how will that be dealt by the app? And if there are legit companies that have different implants, are those represented? Yeah, no, great question. The knockoffs we don't get involved in because there's too many, to be frank. When we talk about companies in India, India is very different in how our where my parents were both born in India. I was born in the States, but very familiar with the way that we do things in India. And so we don't have product patents for a lot of these things. Oftentimes it's just process patents. And, and so you— there could be 10 different companies that have the Karai stem, for instance, that looks identical to it. So we don't, we don't get involved in that. We have, we basically cover all of the, the big US and European-based companies and how they and their versions of these implants. And so if, and that's why one of the things that users ideally when they're posting something on the community feed, they're saying, okay, this was done in India 10 years ago, and then someone responding can say, okay, this is what it looks like, what will probably work, but we don't know what company it is because it could be any company. We just don't know. Yeah, we're focusing more on the bigger companies and not the sort of ancillary ones.

Makes sense.

Kyle?

So I don't know if you have to monetize it through users and user fees. If you build up 100,000 downloads and 20,000 regular monthly users, Somebody's going to want to own that so that they can give the suggestion of how to treat it, right? Oh, you're revising an Accolade stem? Well, you can put in a Reclaim stem with a Gripshin TyFoam acetabular cup and you can use the Diffuse Synthese extraction system, right? So there you have a different business opportunity than just offering software that allows you to pay for access to data or access to new AI. You're potentially building the value of a community that is looking for advice that then can be given extra advice. Now, there may be a subscription model that stems from that. Like, we all have the apps where you only pay for it because paying for it keeps it from giving you advertisements. So maybe there's that side of things as well, but access to being able to recommend what to do could be your most valuable avenue.

Exactly right, and I think you nailed it. And I think, um, that's always kind of been the long-term vision is that, uh, at some point when this is sort of the go-to identification tool for revisions, um, companies are likely going to be interested in, uh, in having direct access to the surgeons that are revising their implants and revising any implants. Um, even— I mean, this could even go for some of the, um, sort of smaller mid-sized companies that, that still spend a pretty significant amount on their marketing budgets and may realize, you know what, this is something that if we give this tool exclusively to our surgeons and our reps, then there's some incentive there for people to switch to these companies.

Yeah.

Well, and you could parse it out, sell the shoulder part of it to one company, sell the knee part of it to a different company. And well, if you want, I don't have all the contacts, but if you want some introductions to some of the like tech-focused, like investment acquisition people at J&J MedTech or ZB or some of the others, let me know and I'll see if you want to meet with anybody.

Yeah, thanks a lot, Kyle. I'll definitely send you an email. That's— yeah, those are the types of people I think that would have some interest in something like this.

Yep.

Good.

Yeah.

Really appreciate it. Raymond, on that exact same note, two other companies come to mind that you may or may not have thought of in case you're looking for an exit. One is Medical Metrics, and I have no idea if either of these make sense, but they're separate entities that may not have as much motive as what Kyle just said, but I know that Medical Metrics is doing some interesting things with spinal motion assessments with, I think, just plain X-rays. I could be wrong. The other company is also way out of the box thinking, and that's the publisher Elsevier. And I know they acquired a virtual training kind of module, and I'm wondering if Elsevier as a publisher of a bunch of medical publications might be interested in something like this. So just given that as ideas. Yeah, no, that's a great thought. Elsevier is an enormous company and has their sort of feet dipped in so many different things.

Mm-hmm. Interesting. Well, this is fascinating. So we've got like 5 minutes and I, and So I, Parth wanted to talk a little bit about AI and augmented reality and he's a spine surgeon. So I mean, you know, Raymond's the CEO of a spine startup. So yeah, tell us where you think it's, tell us where you think it's going and we'll just talk about spine first. Where is AI and augmented reality going in spine?

Yeah, so I think it's going to be able to do the surgeries much sooner than we think. And as far as the question of when are people going to feel comfortable letting a robot do a procedure, that's another question.

Oh, we're losing.

And some people may know that the GBA, um, just a nitpicky cruise that every company probably robot, uh, again, this is, we're talking about the big companies that have robots where the robot can, uh, uh, is that me? You're, yeah, you're a little jumpy, Parth, but where the robots can use AI to put pedicle screws safely. I'll wait a second till it, maybe it may get a little better.

You're good.

It's a little bit better. Yeah. Yeah, so I think, so we're already at the point where all of the companies that have robots have software in there internally that can put screws without any, essentially any human input. Right now, the first thing I think we're going to see in robotics is which the companies are coming out with are robotic decompression. So this usually involves a combination of the robot with a newer sort of oscillating or, or like MySonics type hypersonic saw system. And in conjunction with the robotic arm to do a decompression. This quickly will come to the point where AI starts doing that, where then AI can do the decompression. And then, and I mean, at that point, the question, there's going to be ethical questions, there's going to be questions of safety. But one of the big questions is going to be, are the surgeons actually better than these machines in a perfect world where we know that the calibration is right, we know that the navigation is on? We know these companies are going to come out with systems that can adjust for movements in the bed, which we're already seeing that in the arthroplasty space. So yeah, I mean, there's— I mean, this could be an hour-long just topic in itself of when— of just the ethics of all of this and how we're going to see it play out. But I, for one, am very excited about it. And I think ultimately it's, it's gonna bring, lead to better outcomes for our patients and, and safer surgeries for our patients.

So pedicle screw placement today, decompression is sort of the next step up of, of the technology.

Interesting.

Yeah.

So, so for, for example, Globus bought a company called or a product called the DuraPro, which is an oscillating drill that if you see videos of it, you can drill just right adjacent to the dura without concerns for tying anything up or harming anything. And so when you attach this drill to the Globot, it's gonna be able to safely drill away the bone similar to say the Mako for a total knee. So yeah, I mean, there, and there are gonna be questions, especially for the younger surgeons, particularly the generation of surgeons after mine, which I still was trained on freehand technique and how to do everything without navigation and without robotics. But the generation after mine, we're seeing, I mean, more pedicle screws in 2025 were placed with robotics or navigation than freehand pedicle screws. As that trend continues, I mean, we're gonna wonder, like, are these new surgeons coming out capable of doing these cases without their— without technology as a crutch? But, and then the other side of that is we know that this technology is safe and can make for safer outcomes for a patient. So is— are we— are we being negligent by not utilizing them? Just some of the— some of the kind of things to think about. And I know we're, we don't have time to delve into these, all these topics today, but maybe, maybe another day.

Yeah, we'll bring you back for AI.

Go ahead, Mark.

There's a meeting tomorrow in San Francisco called DocSF headed by Stefano Bini, orthopedic surgeon at Stanford. It's basically the whole subject is AI and orthopedics. Uh, one of my partners is out there. Um, there's a lot of interesting people there, movers and shakers. Yeah, you're on to something there. That's great.

Um, oh, that's—

yeah, we could talk about this easily another hour. In fact, I— as you're talking, I'm wondering who's going to have the longer career, the physician bioethicist The AI ethicist or the physician surgeon?

Yeah, yeah, it could go either way. I think, I think that'll depend on if the ethicists decide if, if the AI can take the surgeon's jobs or not. Right.

Well, this has been terrific. Well, we're gonna bring you back for an AI discussion. Yeah, this is, Every time, as soon as I saw the implant identifier, I'm going, wow, why didn't I do that? You started it 6 years ago and it's an incredible tool and it's just going to get better. So yeah, we're pulling for you. Everybody download the app and play with it. Well, thanks for taking time, especially from Germany.

Thanks, Parth. Thank you so much, Tiger. I really enjoyed it. And yeah, we'll definitely do this again.

Yeah.

And thanks everyone. And yeah, please, my email, Tiger has my email and—

It's in the chat.

So yeah. It's just parth@implantidentifier.app. So yeah, feel free to reach out and I'd love to—

Have fun at Oktoberfest. You have 3 more days.

Yep. Oh yeah, we're, yeah, we got my residency class is flying in from Tennessee and New Orleans. So we're, the three of us are, are meeting up with our families here, so it's gonna be great.

Yeah, go to the Lowenbraut tent, but stay, stay away from the Australian table. They're pretty rowdy.

They are, they get a little rowdy. Okay, good to know. Good tip. All righty, well, hey, thank y'all so much.

Thanks, good stuff.