Perimeter Medical Imaging AI is a medical technology company driven to transform cancer surgery with ultra-high-resolution, real-time advanced imaging tools to address areas of high unmet medical needs. I'm very pleased to welcome CEO Adrian Mendes and Perimeter board member Anantha Kancherla to discuss the company's leading-edge AI technology and medical imaging systems. Both speakers have extensive world-class experience in developing leading artificial intelligence systems at some of the world's top tech companies. Today, we're going to be discussing the role of AI in healthcare and specifically how Perimeter is developing its proprietary OCT imaging technology coupled with AI with the aim of improving patient outcomes and reducing healthcare costs. Gentlemen, thank you very much for joining us. We're going to start off with a short presentation from Adrian giving an overview of Perimeter. Adrian. Thank you, Martin. Yeah, we'll go through a few slides here to introduce you to Perimeter and introduce you to Anantha and me. Just a reminder that we may make some forward-looking statements throughout this presentation. I joined Perimeter about nine months ago as the Chief Executive Officer here, after I'd had about 25 years in technology down here in Silicon Valley. I started my career at a company called Cypress Semiconductor, where I managed various functions. Most recently, before I joined Perimeter, I was a Chief Operating Officer at an AI hardware and solutions startup called Groq, Inc., where I worked for about seven years. At that company, we grew that company. It was a startup. There were a handful of engineers and employees there when I joined. By the time I had left earlier in 2023, we had grown that company to about 250 people, we had raised a significant amount of money, we had hit a billion-dollar valuation on our latest raise, and we had product and market, and we're working on a second-generation market. When I was introduced to Perimeter, with this opportunity, I was extremely excited, because this would be an opportunity to take what I had learned over the past seven years in the AI industry and then apply it in a brand-new area in a place that's much closer to the end users, in a medical field, which allows me to take what I've learned with AI, and then bring it into a whole different area of the economy and of the industry that was helpful then to a whole brand-new set of users. That's my story. With that, let me hand it over to Anantha to give a little bit of background on his work experience. Hi, everyone. Great to talk to you and Adrian and Martin as well. My story starts like in India. Like, I went to IIT Kharagpur, where, actually, I worked on the last generation of AI. When I came to the U.S. to pursue further studies, I was told that AI was dead; that was mostly because, like, the last wave of AI died, and we were in this what was called the AI winter. All of us started working on alternative things to do what we call IA, which is intelligence amplification. The idea was that people are pretty smart; why not just build tools for them to, like, do a lot more with what they have to make them even better? I worked on graphics, and ironically, there is a connection with what I did. I was at Microsoft and worked for many years with companies like NVIDIA and AMD, so now you know what's coming. We worked on the GPU architectures back in the day. Today, like fast-forward, this is really where this is —the hardware on which, like, AI is happening. Along the way, somehow I reconnected back to the AI world. First, when I was the VP of engineering leading the software at Lyft's Level 5 organization. We were building a robotaxi, which is the most and hardest problem that one can solve with AI. So it was a lot of fun and a lot of learning, understanding the limits of where the technology is at. That's when I really reconnected with AI. Then I went on to Meta, where I led its AI platform team. Meta, as you know, like, has many products, all of which use AI. Whether it is the ranking in News Feed or whether it is like the head-mounted display to do virtual and augmented reality, all of them use AI on different scales. The scale is enormous. I also got to, like, work with people who actually invented AI back in the day, back in, like, the early 2000s. Very, very recently, I joined General Motors to lead their ADAS organization. ADAS is automated driving and advanced safety. Again, as you can imagine, we use every high-tech AI technique possible to make driving safe and keep people safe on the road. Thanks, Anantha. Okay, let me flip through our about page for our company a little bit. This right here is our mission. You know, you can read it here, but, you know, we envision a world where patients no longer experience the emotional and physical trauma of being called back for a second surgery due to cancer left behind. This is obviously a big mission. It's one that sort of drives the company, drives all our employees, and drives our board of directors to help solve this problem. We actually believe that we've got the technology to do that. The combination not just of the AI, but really that overlaid on top of a somewhat novel imaging technology—those two things together are what we believe are going to allow us to do this. I think we're well on our way there. This is our mission; this is what we're working towards. On the left side of the screen is a picture of our product. It's a cart device that sits inside the operating room. What you're seeing here is what is on the market right now with FDA clearance. It's called the S-Series OCT device. What this uses is a near-infrared light source that allows penetration of tissue to a millimeter depth. With that, what it helps surgeons do is visualize the microstructures that are within those first couple of millimeters. Why that's important is you can see with the OCT technology, this optical coherence tomography technology, the differences between different types of tissue. What's key to this technology at the OCT level is the resolution. As you can see written here, OCT with this near-infrared light source has much higher resolution than both X-ray and MRI. The images that you see on the right-hand screen here kind of show what the surgeon can see when they use OCT, which is the image on the top, and then how that compares to what the pathologist sees later on, which is the image on the bottom. Really, the difference is that the pathology, you know, the pathology data comes back to the surgeon a week later after the patient's already gone home. Now, when we have the device in the operating room and the surgeon can see this in real time, the patient's still there, so if they observe something that looks suspicious, they can look at that and then make a clinical decision while the patient's still on the table to do what they call a shave. What I've just described is our base technology, the OCT technology. Nothing here has to do with AI. Really where the power comes in is when you can overlay some AI on top of that. What you're seeing here on the picture on the right side is an image of our B-Series device. This is what's in clinical trial right now, not FDA cleared; we're going through the trial to get it cleared. We should be finished with the trial later this year. What the AI does is it has an image recognition algorithm that reads all those images that are taken. We've trained on those images. We have a database, a very, very large database, with over two million images that we've collected. We've trained our algorithm on that. The algorithm can go through, look at all those images, and highlight those images that are most suspicious that there's potential for some cancer to be in the margin. That allows the surgeon to flip through those, a very small subset of images, when they're looking, and it helps speed up the workflow. Think of it as sort of a, you know, as a co-pilot or an assistant for the surgeon as they're looking through the, you know, everything that was scanned to be able to really look at the dozen or so highlighted, you know, most suspicious areas. This helps speed up time in the operating room, helps bring confidence to the surgeon that they're not missing something, helps improve the usability of this technology for, you know, many more surgeons, and really helps make this device, you know, take it to the next level of usefulness. Right now this clinical trial is ongoing. Here are some of the, you know, the hospitals that are using it. Dr. Alastair Thompson is our lead PI, a lead investigator working on this. I think this is, you know, we're very excited in the company about getting the results of this trial later this year. As that trial progresses, and if we're successful, you know, with what we're trying to do there, this will allow us really to bring this, you know, this AI algorithm, into the marketplace and then provide it to our surgeons to allow them to benefit their care, you know, their patient care, for the patient they have coming through their operating room. You know, that's it for the presentation part of this section. Hopefully that gives you some understanding of what we're doing from a base technology and then the AI —how the AI is helping our customers use that technology even more. All right. Thank you. Both of you have exceptional tech and AI resumes. You went over your histories, but with this kind of resume and this kind of market for AI, you could have chosen pretty much to go anywhere and to lend your expertise. What was it specifically about Perimeter that drew you to this company and this opportunity? Yeah, great question, Martin. Maybe I'll start. I'll give my answer, and then I'll hand it over to Anantha. For me, when I was introduced to this company, my first reaction to it was that I don't have a lot of experience in the medical device space and medical imaging. But I do have significant experience in the AI side, you know, of the house, for my last seven years at Groq. Really, when I started at that company, AI was just starting to find its footing. I've seen the industry develop over the last six or seven years. Through that, what I've observed is. When I started at Groq, really there were very few companies that were taking advantage of AI, and it was really the high -tech companies, the Googles, Facebooks, Netflixes of the world, and and Microsofts. As the technology has matured, as the hardware has become better, as the tools enabling folks, engineers really, to develop have become better, what I was getting more and more excited about over the years was the ability for this to get propagated beyond a small set of companies, high -tech companies. This was already, you know, this idea of AI in other industries was already sort of on my mind. When I got introduced to Perimeter, I was primed in that way, so I think the next phase of AI evolution is going to be able to bring this to kind of the everyday people, outside of the tech world. Specifically with Perimeter, what I was very attracted to was the fact that there is an imaging technology called OCT that Perimeter uses that's actually in widespread usage for not the application we're using, but mostly for ophthalmology and some vascular applications. But this was the opportunity to take an imaging modality and apply it to a brand new area, number one. Number two, the company had a lot of patents around some unique applications, you know, unique ways of using OCT that allow us to really carve out a niche for ourselves that's fairly well-protected. The third thing was the fact that the company had already built up a pretty large data library, a dataset, which, for me, in my experience, is very attractive because that's actually where the value you see for AI comes in: the ability to have proprietary datasets, the ability to train models, and then to, you know, use that in a way that is very hard for other companies to replicate. I think the last thing I'll just kind of say before I hand to Anantha is the team was very strong. Anantha was on the board, you know, before I joined. I was very impressed with Anantha's resume. I looked at the team internally at the company and saw what their resumes were and what they've been capable of producing both here and in other places, and it made me very excited about joining a world-class AI team, you know, layered on top of a world-class medical imaging team. Yeah, I'll go next. I'll go next. Yeah, I joined Perimeter like maybe nine months before Adrian did. Well, you know, like, I've always been fascinated by this AI thing ever since I was an undergrad a few decades ago. In the, you know, early part of my career, it had just kind of faded away because the technology didn't live up to its promise. When it came back in the early 2000s, early 2010s, actually, using the same stuff that I was working on, in the interim, it just kind of felt like an absolutely natural thing for me to gravitate towards. Once I kind of started working on it. Like I said, I worked on a really hard problem first, which is self-driving cars. That's where you deploy every possible imaging technique and try to understand what the world looks like. You have to use AI to, like, really understand what the surroundings are and what the machine is able to see. That gave me like a view. That was a really hard problem; it also gave me a view into what modern AI is capable of. It's, like, very similar to Adrian; I was also thinking in terms of, Oh, wow, if we can do this, what else could we be doing, and how else could it be beneficial to mankind? There are so many different places where we could be employing it. Like, you know, like all the places where we really need help. I was thinking about the spaces where it could be done, and, you know, like, healthcare and climate; there are a number of different places, and all of them are, like, very, very worthy of, like, benefiting from AI. When this connection with Perimeter happened, and I saw, like, this is something that can genuinely help and improve people's lives, and that 's for people who badly need help, it was a no-brainer for me to say, Okay, how can I help this company? How can I help them, like, to use AI to transform? It did help that, like, there were very smart people who knew what they were doing. This OCT thing was brand new, and being familiar with, like, lidars and cameras from my self-driving car world, it just kind of felt like, Oh, yeah, okay, this is just yet another modality of imaging, and AI should easily be applicable to that. It did help that Perimeter has, like, a pretty good proprietary dataset that could be leveraged to build some pretty cutting-edge applications here. That's basically why I came here, to help Perimeter. All right. Obviously, with especially Adrian bringing on someone with your AI expertise into the leadership role at Perimeter, AI image recognition is a key aspect of the company's technological path going forward. The key benefits from AI with the imaging system—can you elaborate on that? I guess with that, as you said, the surgeon is trying to identify when all the cancer is out of the patient so they don't have to cut out too much cancer and they also don't have to worry about having to go back and do another surgery. That reading of the image is very important. AI is being used here to augment the surgeon. I guess, could you? The benefits would be, A, to let the surgeon make better decisions, where the AI helps them identify what is and what is not tumorous or cancerous tissue in there. I guess to speed up the whole process. Can you elaborate on how the AI benefits the surgeon and ultimately benefits the patient, most importantly? Yeah, absolutely. I think, you know, let's start with focusing on the Image recognition piece of it, which we've talked mostly about. If you think about sort of the situation a surgeon's in, it's a very highly stressful situation. They need to understand what's going on inside the operating room in many different dimensions, and they've got a patient, you know, that's under on the table. There are a lot of things vying for the surgeon's attention. Now there's this technology that helps them make sure, exactly as you said, they're not cutting out too much tissue, but they are getting all of the cancer out, which is exactly what the patient wants. The easier we can make that for the surgeon, the better. What the AI does is it does a couple things. One is it gives that surgeon confidence that they are seeing everything that has been imaged and paying the most attention to the parts of the image that are the most suspicious. If you think about it, you know, when you image a tumor, there are going to be hundreds and hundreds of images that are being captured from the OCT device, from our device. If the surgeon has to look through all of them, they can flip through them kind of like a movie. They can flip through the whole volume. It takes that amount of time to kind of concentrate and make sure that they're catching everything. From their standpoint, it would be great if they had an assistant saying, Hey, look here, look there, look in this other place, 'cause these are the areas that are most likely to be the places where there might be some cancer in the margins. Great. What does that do for the surgeon? It takes a mental load off of them or reduces the mental load while they're in the operating room. That's number one, making that job easier. Number two, it helps speed up. If the surgeon has to go through the entire volume, that's going to take a certain amount of time. If they have to zoom in to, let's say, 12 images only, well, that's going to be much quicker. For them, they're really trying to get the, you know, get the job done, get the patient closed, and then send them off to recovery as quickly as possible. These are the two main areas from an Image recognition standpoint; this AI is going to help that surgeon. I think there's an element of it also, which plays to our Market from a business standpoint, our market expansion, where there's going to be some surgeons that are going to be more willing to adopt the technology once there is an assist, an AI assist there that's going to help them do their job a little bit more easily and get them more comfortable with it quicker. For me, getting this next, you know, the B-Series, the next -generation device, cleared by the FDA and onto the market is going to help us with our market expansion quite significantly, I think. I think that's it from the Image recognition standpoint from the surgeon. There are other areas where we are using AI deeper inside the technology stack to help speed up the image capture and to help improve the clarity of the images and the quality of the images that we haven't really touched on. You know, our AI team's kind of very active all the way up and down and through the stack. One of the areas in AI that has benefited in many of its applications across industries is, as you hinted at, the skill gap reduction. Where, when there's an excellent radiologist or a surgeon who has a lot of skill and practice in analyzing the images, they could do a better job at it. Where if there's a newer surgeon or a surgeon who doesn't have the same number of surgeries to go through, this can help increase their expertise and their confidence as well. Is that one of the sort of improvements that the AI adds to the process? Absolutely. If you think about it, think about, you know, America, right? There are many people who live close to a metropolitan center, where there are surgeons there that do hundreds of procedures a year, hundreds of lumpectomies, a year. There are about 8,000 surgeons that are doing at least, you know, one or two lumpectomies a year. If you live near Dallas or San Francisco, New York, or Boston, you probably have access to those surgeons, and those surgeons are going to be well-practiced and at the top of their game. There are also a lot of people in this country that live in more rural areas where they don't have access to those, to those types of high-volume surgeons. The surgeons that are helping those patients are not going to have that type of volume. Part of our goal is actually to be able to bring this tool out to those lower -volume surgeons that don't get the practice, you know, that maybe aren't getting as many, you know, reps, so to speak, in lumpectomies, and be able to bring these tools out to them to help them actually get the same results that the high-volume surgeons are getting. This AI really helps with that, where it helps allow a surgeon that's only doing, let's say, 20, 30, or 40 surgeries a year, lumpectomies a year, get up to closer to the skill level in sort of assessing the margins as those surgeons that are doing hundreds per year. We see this as being a way to help that. Surgeons missing some cancer within the patient isn't a rare phenomenon. There is a significant re-incision or re-operation rate. Could you dig into that and what potential benefits the technology you're doing can to improve that? Yeah, absolutely. Right now for lumpectomies, the re-excision rate is on the order of 20%. If you think about that, 200,000 lumpectomies a year in the United States—20% of those patients have to come back again for a re-excision. You know, if you run the math on that, it's 40,000 re-excision surgeries per year in a situation where, unfortunately, if the surgeon had been able to understand those margins and be able to see that in real time, those are 40,000 surgeries that would not have to happen. 40,000 women that don't have to get that phone call a week later and say, you know, and hear that, Oh, I'm sorry, we thought we got it all out, but we didn't. 40,000 phone calls surgeons don't need to make. It's a pretty, it's a pretty big deal. Now, our goal is to try to reduce that and try to reduce that significantly. You know, we're starting to see some very interesting results from the surgeons who have this, you know, have our machines in the field. Also, many women get a full mastectomy where the entire breast is removed for fear that some of the cancerous tissue is left behind and could metastasize and spread to other parts of the body. Does it have the potential, then, to give comfort to both the surgeon and the patient that the full mastectomy is not required and that just a less invasive or less intrusive surgery is a viable and safe option? Yeah. Yeah. You can imagine yourself, you know, in the situation of being a person being told that you've got cancer, and the only thing you want is to make sure that you don't have cancer after, you know, after everything's said and done. There, you know, there is a portion of women that will say, You know what? Just let's do the mastectomy, and that way we can have the highest chance of not having a recurrence. Now, of course, you know, if you speak to women that have gone through this, the vast majority of them actually want to preserve, you know, as much of their breast as possible, and so they choose lumpectomies. They do have to face that trade-off of 20% of the time that there is a re-excision. That's a very, very hard decision for a woman to make. Our goal is to try to help reduce that probability of having a re-excision to help make that decision, you know, of the patient and their doctor much easier. The other thing I do want to highlight, Martin, is this technology's not just applicable to breast cancer. This isn't just women who are facing breast cancer. Although we've spent so much time talking about that, this same problem exists across pretty much all cancer types. 20% is the number for lumpectomies, but it's in the teens and higher across prostate, head and, you know, head and neck cancers, and colon cancer. This is something that, you know, not everyone, but a lot of people will face throughout their lifetime. The goal, if we can help surgeons reduce those re-excision rates, will really help a large, you know, portion of the population who have to face these decisions. Reading the headlines of the world of AI and how it's evolved so rapidly over the past 18 months since ChatGPT first burst onto the scene, what people have learned is that it's not just the algorithm; you have to train the algorithm with good knowledge and good information. You referred to the fact that you've got over two million images to train your algorithms with and your whole tech stack with. That seems pretty large and impressive. Could you just elaborate a bit on that, the significance of that database of knowledge? I presume it's a growing database as you do additional surgeries and your algorithms improve, so the accuracy and quality of your decision-making improve. Yeah. Maybe I'll just start, and I think Anantha Kancherla probably has something to say about it as well. We've collected data through, you know, obviously, many different tissue types, both with cancer and healthy tissue. When I refer to two million, I'm actually just referring only to the breast cancer —or sorry, the breast tissue —data we've collected. We've actually more across other tissue types. Those images are created in multiple ways. Every time we scan, you know, every time we scan a tumor, we actually have multiple images of that, so that goes into the database. There are ways to augment the images that we have, so that adds to the database. All of this just increases the intelligence of not only the algorithms we've got but also opens up the doors for us to create even more sophisticated algorithms without fear of overfitting and things like that. It's key to any training, any type of AI. I think that's pretty common knowledge. We do have a very strong focus on trying to grow that database continuously. Yeah. I want to add that, like, it's more than just simply the image recognition work that is happening at Perimeter. Even like the one that you mentioned to denoise, that's actually like a very clever and pretty cutting-edge application of AI. It actually uses a model called the diffusion model. If you've heard about generative AI these days, I don't know if you've had a chance to use something like Midjourney or something like that, or DALL-E, like they create images. They use something called it. It's a technique called Stable Diffusion. What we use here in Perimeter is very similar to that. Instead of like generating new images, we generate clean images without the noise; that's basically like how the surgeon is able to really clearly see without actually like cranking up the resolution or taking double or four times the amount of time to scan. Everything that you're doing in a different part of the stacks is like very AI related. You mentioned also like data augmentation. There are so many other things that are going on inside Perimeter that push beyond the traditional Image recognition side of AI. Yeah. I think I'll just, you know, add to that, Martin, in that this. You know, when we talk about the attraction for me to the company, and then what keeps me excited about working here, is that the AI. Well, both the AI team and what we have, and what we continue to grow, as well as the different ways we can use it to help the business, are multidimensional. It isn't just an image recognition algorithm, and we've got a team to totally focus on that. Actually, the team's able to, and then they continuously do, look in the world of development, you know, the AI world of what's being developed out there, from an algorithmic standpoint and a technique standpoint, and then think super creatively about how to use that across all elements of the business. When you've got a tech, and I've observed this, you know, since I've been involved in AI since 2016, you sort of have this green field right now for brand-new toolset AI and brand-new technology continuously getting better. What limits, sort of what the limit is right now, is the creativity of the team of how to apply these techniques and tools to various parts of the business. This group that was built even before I joined here and that we continue to evolve is just very, very good at that, both on the classic sort of image recognition piece of it, which is the most easily observable part of what we do, and even, you know, under the hood in many, many different, very interesting ways, like Anantha alluded to when he spoke. You've talked about the different levels within your AI stack, and you have a clear objective and purpose with your current technology. It sounds like it could be used for other sorts of images where you're cleaning up the image and identifying it. Is it limited, this technology you're developing, to the OCT platform, or are there other modalities like X-rays or MRIs or, I don't know, whatever other types of imaging sources that this technology could be used for as well? Yes. It's absolutely not limited to the OCT, you know, technology imaging modality. It's an interesting question. If you think about it, you know, cancer in the whole sort of, you know, diagnosis and treatment of cancer, there are lots of different images that are taken. There's X-ray, you know. mammograms are a type of X-ray, so there's X-ray imaging. There's OCT imaging. There could be different modalities, like ultrasound or MRI. All of these put together would make for a very interesting data set for, you know, layering AI on top of. What we're very, very careful of here is as we think about how we evolve the technology, not just making sure we don't back ourselves into a corner where it's really only really good for OCT, but really making sure we've got pathways open that allow us to take different data sets in various libraries and then train models that are multimodal. I think there's a lot of power you can get in terms of helping surgeons and helping patients in their diagnosis and treatment if you can pull some of this together. Yeah. Maybe I can also add an analogy here? Like, I mean, my background is like working on autonomous vehicles; the way these vehicles work is that they use like different types of imaging technologies. They use radar, they use lidar, they use cameras, and they see the same object with these multiple different modalities. What AI does is that it can even learn if some things are easier to see in an image, and then you know it's the same thing that is seen in the radar. In the night, your camera might not be working as well; then maybe you can identify it with the radar. You can actually transfer the learning from one to another. You see some similar things, like when these models are able to understand the structure of a language, whether it is French or German or Spanish or English, and they are able to like figure things out based on that. This technology is incredibly powerful, so it just crosses, cuts across. It's trying to really understand the underlying patterns, and it goes deeper than just simply the imaging modality that is there. The potential is like vast and to be able to make a multimodal system that is able to learn from one modality to another and then take it back and try to interpret things in a much more deeper way. I would think also we've, or many of us have, seen the AI-generated videos now and generated images of cats or whatever. When you look at them, there are always some glitches. There's something you look at and say, Oh, there's something a little off here. You're operating in an environment that's FDA regulated, where you can't have just some sort of random bad facts thrown into your images or your analysis. You're working with the surgeon; there are sort of two sets of eyes on the image. I would think that developing a model under a highly regulated and sort of limited. You're not allowed to make mistakes, so to speak. Kind of like the car driving situation. You can't run through a red light. Could you just discuss a little bit how under a sort of a heightened risk or a zero -tolerance environment, the AI process is different than if you could be sloppy in making cat images or videos? Yeah. Anantha, why don't you take that? Because you've had some experience with this. No, this is a great question. I think this is why I'm, like, drawn to Perimeter because it's so similar to the kind of work that I do because you cannot make a mistake when you are putting a car on the road. Similarly, you should not make a mistake or cannot make a mistake in a diagnosis or healthcare. It's very, very similar. I think that the way these things work is that actually most fields, AI is not yet there at a point where it is human level, which means it cannot do, like, everything that a human does yet. It will maybe be down the road, but that's like in the future. Today, the best way to apply AI is as a co-pilot. It is basically working hand in hand with the human, and it is actually like making the people much more efficient and faster. Where it works very well is that, like, people tend to lose focus, and we don't have the level of energy or speed to look through a huge amount of data and datasets. What AI is able to do is it's able to narrow things down. It can actually do those things very well. Where it can't do very well is like, deal with things that it has never seen before, and this is where humans are very good at. This combination of, like, putting an AI and a human together, like in the case of Perimeter, we are doing the surgeon and the device together, actually is the right way of doing it. This is where, like, you'll see the most success in applying AI. You didn't want to add anything? I think that's it. We're very careful about ensuring that part of the reason we're having such a big image library is to be able to train a model up to a very high degree of accuracy. The latest one, and we just published a paper, demonstrated over 98% accuracy. There is an element of it where you don't want it making stuff up, of course. The good thing about our application, to Anantha's point, is that it operates as a co-pilot. From a business standpoint, if we don't have a high degree of accuracy, the customers are not going to adopt it. Before it even has the ability to create a problem for a patient, we need to meet an even higher bar, which is the surgeons, or else they're not gonna adopt it. The good news is we are getting the adoption on the device before the AI, you know, is available on the S-Series. You know, we're seeing great market traction on that front. Through the clinical trial, you know, we haven't seen the results of that yet, but we do know that the AI itself is up above 98%. When we do bring that to market, you know, upon FDA approval, you know, we have high confidence it'll get adopted. If it doesn't, we'll know what we need to work on. That's sort of the quality filter, both between ourselves and the surgeon, to make sure that we don't impact, you know, patient care. I would imagine AI is new for the FDA as well, and they're trying to wrap their brains around it, how best to apply it to increase the skill level of the surgeon and the efficiency, which can have better outcomes and lower costs and greater efficiencies throughout the whole healthcare system. I would think just you're on the cutting edge of this FDA approval system as well; I presume that the knowledge you're building from that is applicable to many new modes or modalities as you've discussed as well, and that you're creating this real knowledge base of not just the AI, but how to apply it to real -world life and death, literally life and death situations. Yeah. Yeah. That's right. We maintain a good relationship. We have a good relationship with the FDA, working through this trial especially. They're learning how to regulate technology like this. I think it's very important what they do. You know, it is brand -new technology. You want to be able to bring to bear on, you know, for patient care, all the best, newest technology, but you don't want to do it in a way that's sloppy. I think the FDA is doing a pretty good job balancing these two. They know what they know, they know what we're still learning as an industry, that we have to go through these cycles. Like every other regulated industry with new technology, you know, we'll figure it out. You know, we understand this. I think the FDA has been very, has been a great partner with us through this. We'll continue working through it with them. Excellent. We've covered a lot of topics here, gentlemen. I really appreciate your time. Is there anything you want to highlight or emphasize or anything we missed in this discussion before we wrap things up? Actually, I want to add, going back to the point that we were talking about earlier, like, about AI being able to replicate the best. I think that is the attraction for me with AI. Like, for example, like when the style of AI that is used is called supervised learning. Where, like, we get a bunch of data, and then we put experts to go and label the data. Basically, we use experts to train the system, the AI itself. Better the experts that we bring in and the more time that they're able to dedicate, the better the system gets. Now that's the brilliance of the system. That means, like, you're able to, like, bring the best expertise to every surgery regardless of, like, the experience of the surgeon. This is true for, like, any application of AI where, like, you can collect the best data, and you'll see that. In the AI world, we always talk about it as if, like, data matters the most. Better your data, better your AI performance. I actually think that this is actually like really great. Like, we can bring the best experience to the most people. That's best for everyone, healthcare outcomes, economics; it just makes the world a little bit better. That's great. That's right. Well, gentlemen, thank you very much. It was fascinating. I learned a lot. It's great, the innovation that AI is bringing to the world of healthcare 'cause sometimes some of the things that the big AI models are doing, I'm not sure if that really benefits anyone. What you're doing is pretty clearly that it can help a lot of people out there. Thank you very much for taking the time. Yeah. Thanks a lot. Thanks, Martin. Appreciate you having us.