KBO Systems

0:02 Hi, Richard here with Benmore Technologies. and I want to talk about our work with KBO Systems. So, what KBO Systems, um, aim to set out and solve was basically to provide, um, functional insight into the level of sedation, um, of someone who is sedated.
0:19 So, for example, um, during a surgical procedure, um, almost creating a speedometer, um, for, you know, someone who is administering sedation, uh, potentially maybe an anesthesiologist, um, to just determine how sedated that person is.
0:37 Are they close to waking up? Are they extremely sedated? Are they somewhere in the middle? Etcetera. Um, this has various, you know, um, benefits. Of course, you want to know if someone is about to wake up during a procedure.
0:51 Uh, and, um, and, uh, if I recall correctly, there's also adverse side effects if someone is extremely sedated during, uh, procedure as well.
1:00 Um, I think that there's, it, it just takes longer for them to wake up after the procedure and, and stuff like that, right?
1:08 So there's, there's benefits to having the speedometer, um, you know, of course monitoring when someone's about to wake up and also not making any sure that they're extremely sedated.
1:17 Um, but in relation to that, um, basically our work and, and, and what we did for them was we basically trained an AI model, um, on a public data set, um, to, uh, replicate, um, this sort of functionality, replicate this, uh, monitoring technology.
1:36 Um, so incredibly interesting. Um, uh, project, but essentially what was encompassed is, um, there is an existing, uh, medical device called the Biz Monitoring System, um, created by Medtronic.
1:54 And this is an existing solution that aims to create, like, almost a speedometer sort of thing, where based off this biz number, um, you know, someone who's in administering sedation, um, can determine how sedated someone is.
2:08 Okay, that being said, the algorithm that comprises this medical device, when it receives EEG waves, which is basically just your, um, electrical waves, um, that can be read by putting a monitor across your forehead, basically it reads these EEG waves, and then there's some, there's an algorithm that
2:28 determines the sedation depth. Now, with many things in software, this is incredibly interesting to learn, if an algorithm is extremely proprietary, often times the best way to protect that is not even to file a patent on it, it's just to keep it as secret within the company.
2:45 So, in relation to this algorithm, it was just kept secret. There's patent filed on other parts of the methodology here, but the core algorithm is just secret.
2:56 Like, no one knows exactly how it works. There is a research study done by Nature, um, and here it is, and this is linked in the case study write-up, but essentially, they were basically doing a similar sort of thing to what we were doing.
3:10 They were trying to understand the biz algorithms. Okay, and if we look over the abstract Essentially, what's going on, right, uhm, as you can see here, only a portion of the proprietary algorithm has been identified, right.
3:26 They investigated the biz algorithm using clinical big data, right, and they used a public data set to determine exactly, just to try and glean some insights into exactly how it works.
3:37 This is an interesting read. But, basically, they're just developing a study on how that algorithm works. So, essentially, what we did is we tried to kind of replicate this study.
3:49 We used the same data set that they did. This is the VitalDB public data set. You can see from this data set what we can get is we can get the input into the algorithm, the EEG waves, and we can get the output, the biz numbers, right, like, whatever that is.
4:04 So, basically, our idea was, well, let's try to glean a little bit more insight into this. Let's try to create an AI model that basically replicates this biz algorithm and see if that provides the KBO system team with any actionable insight.
4:18 So, that's exactly what we were able to do. We took this data set, this public data set, we knew the input, and then we knew what the output was, and we just trained an AI model.
4:28 To basically try to replicate that. Without any knowledge of what the algorithm was doing. As mentioned, this is how it currently works.
4:36 You have a physical monitoring device, you have EEG waves going into the big biz algorithm and it's outputting a biz score.
4:42 So, we just said, well, we have a large data set of EEG waves, the inputs, and we have a large data set of the outputs, the resulting scores.
4:51 And we were, we were able to replicate the algorithm, I think, to an 88 or 90 percent effective rate. Uhm, we split the data set up into a training and testing set.
5:01 Uhm, and this was honestly outside of our realm of expertise. So, we, you know, we contracted and we pulled in a colleague, Stuart, uhm, one of, frankly, the smartest people I've ever met in my life.
5:14 He's currently studying artificial intelligence. Uhm, getting his master's at Duke, uhm, but we pulled him in to help us out with, uhm, the investigation of this, uhm, and development of the model, and, uh, we were able to file a provisional patent on behalf of the KBO Systems teams with our findings
5:32 , and they were, uhm, you know, able to glean some actionable insights. Regardless, across the board, just one really, really, really, really, really cool project to work on.
5:42 Again, you get to learn about the intricacies of medical device monitoring, and you get to, you know, work on, uh, just a very interesting, you know, project in relation to sedation depth. Uhm, lastly, you know, we got to learn a lot about the FDA approval process, uhm, what that entails, some of the
6:01 limitations associated with getting, uhm, you know, software approved. Uhm, and one of the most interesting things that we learned, as mentioned earlier, was like, sometimes in software, something so valuable, you don't even file a patent on it, you just keep it secret.
6:16 Uhm, and going through the patenting process, we knew this a little bit before with previous projects, but basically, uhm, software is extremely hard to patent, it's software.
6:28 And, as mentioned earlier, like, the moment that you file a patent on something, it's just software, like, you can just write it.
6:35 It's just code. So, often times, and this is something that we tell to, you know, prospective clients or clients, anyways, it's just, hey, if, if you really think that you're doing something extremely innovative and research-driven, you probably just want to keep it as a secret.
6:52 So, anyways, that's pretty much just the work that we did with KBO Systems, of course, just a very interesting project overall, and, uh, something that we're very proud of.
7:02 Again, I want to shout out Stuart, genuinely one of the smartest people I've ever come across, uhm, and his work and, uhm, what he was able to help our firm do.
7:12 Uhm, and it was just an incredible experience getting to work with him. Uhm, as well, so.