Hello, everyone. Thank you for joining us again. That was a fantastic first two sessions. And now we are going to go to our third session. I'm Bonnie Conda, and I'm from Baylor, Dallas, and it is my pleasure to be joining Prateek and Moen in the third session on Barrett's esophagus diagnosis and neoplasia. And we will also have Linda and Venkata guide us and coordinate us as they did in the last session. Thank you. Thank you. So Prateek, I think we can hear you. And you can see Moen's ready in action. He's got his jacket off. And so I think he's ready to go there. So Prateek, are you there? Can you hear us? Yes, I can hear you loud and clear. Hello, everyone. Can you hear me? Yes, we can hear you too. So I hear that the patient's just getting prepped and stuff and that you have a short video case to go over with us. Yeah. So while the patient is being turned around, if I can show you the AI system called Wise Vision. We just recorded a couple of cases. So if I run the video and talk you through. So can you run the video, please? Can you see our screen? Yeah, we can. Yes, we can. Oh, good. Can you play, please? So that's a patient with a Barrett's lesion. And you can see that the reverse green L is a sign of detection. As soon as it detects, it transfers the image to the top right corner, and then the heat map shows the location of the lesion. So as we come back now, the lesion is back in focus. You can see the heat map is detecting. So moving on to another case, which is a splat lesion. Again, you can see as soon as they come into view, the reverse L finds the lesion, and the heat map is showing you the location of the lesion. The last case is a very subtle lesion only seen on retroflexion. And you can see that it's very difficult. But even in retroflexion, the system is detecting the lesion. So the heat map is showing you the location. There you can see slightly 2C depressed area. This is, I feel, is the common blind spot for community endoscopists who do big volume endoscopy, and lesions around this area get commonly missed. And this will be one of the big advantages of AI in the future when you have lesions like this, which can be picked up because they can be very easily missed. This is just on top of the gastric fold, and you can see from the heat map that it extends only up to the lesion and not all the way down the gastric fold, because gastric fold is always a nuisance element. And AI can pick that up as a false positive signal all the time. So, these were four very quick snippets of what the system does. What I would ask is our colleague to switch over to the room view to show you how the system is set up, because there are various ways of doing where the system could be integrated into your main screen, which you're looking, or you could have a parallel screen next to your main screen. The system that we are using is basically showing, I guess you can see our room view now, yeah? Yes, yes we can. So, we have the main screen on the left side, which is the endoscopy screen of the endoscopy stack, and then on the right side we have a second trolley here with its own screen. This is the AI system. They stand right next to each other, and endoscopists can decide to ignore the main screen or focus on that main screen and look at this side screen only when there is an audible alarm. So, whenever there's a detection, there's an audible alarm, and they can look at it. So, there's a pros and cons of having two screens, because you then have to change from one screen to the other. At the same time, it avoids all kinds of nuisance because of false positive detections which can happen with AI. So, hopefully we'll be able to get the endoscopy in very quickly. Before we do that, I'll give you the history of this patient, who is a 75 year old gentleman. About 18 months ago, he was diagnosed to have a T2 adenocarcinoma, which on histology showed poorly differentiated cancer at the center, which is about 80 miles away from us. He underwent radical chemo radiotherapy, and followed by surgery. So, he had radical chemo-rad isoflagectomy, and at the time of surgery, they did the donut test, and they found dysplasia at the margins. With the COVID, he didn't have further follow up, and he's now had a further endoscopic assessment, and he's got residual barrets, and he's got dysplasia in this residual barret right next to the anastomosis. That's how he's been sent to us from the referring center, where they did not see a visible lesion, but they took biopsies of residual barrets, and the biopsies have shown high grade dysplasia plus low grade dysplasia. So, this is a big challenge for the system now to see if the system can see, or for me as well, whether we can see. So, let's switch over to endoscopy view in a minute. Hey Pradeep. Yeah, go ahead. So, this is the Fuji system? So, we're using a Fuji endoscope, but the system is from a company called NEC from Japan, National Electronic Corporation, which is a big AI company, not in the medical world, but they are well-established in artificial intelligence. So, they are just developing this AI system for medical care now. Okay, so just enlighten us here. Are we using this AI system to detect barrets or to detect the neoplasia? That's a very good question. When we started developing this process, that was the first question asked to us, is that should we develop AI for detection of barrets? And I said, well, barrets is mostly a Western disease. And in the West, most people know what barrets look like. The challenge is finding neoplasia in the barrets. So, no, the system does not detect barrets, it only detects neoplasia within barrets. And does the system focus on the texture it sees, the patterns, the morphology, a combination? What are the drivers for it to light up on that heat map? That's a very, very good question. I ask that all the time to the engineering team to say why exactly it is seeing. Is it seeing pit patterns, vessel patterns? And they say none of those things. It looks at pixels and pixel densities and texture of the image and color and contrast. So lots of things. The true answer is I don't know what it looks at. And it doesn't really follow the way we detect. But what we detect is also visual clues of change in color, change in texture. And I think AI is doing similar thing, but true answer is I don't know. Can we go to the endo image, please? Give us the endo image. So can we move to endoscopy image, please? And then Pradeep, before you switch on to AI, perhaps if you can just walk us through also with your HD Fuji scope, if you could maybe also show us some BLI images and to see how the barrets and the G-junction post-surgery looks like. That's a very good idea, Pradeep. Unfortunately, our transmission is connected straight to the AI trolley. So if I can do the reverse ray, and then I'll do, because you can see that looks to me like the lesion at six o'clock. If you can come back. So that's the squamous. And you can see now there's no positive signal, there's no alarm. And if we go down slowly now towards the barret, and so far there is no AI signal for a detection. And here we come, you see the reverse L appears and it transfers an image to the top right corner with the heat map showing a lesion at the bottom. So the AI is detecting a lesion. I also agree that there is likely to be a neoplastic lesion here between about six to eight o'clock. Looks slightly depressed. Pradeep, doesn't it appear to be a little bit more extensive than six to eight? I mean, how about it's almost starting at the three o'clock position. Well, yes, that's where we will have to use the BLI. And I think you're probably right. There's a subtle extension at about three o'clock. So it's almost three to nine o'clock. Extensive. And Pradeep, this is a good example also is that a lot of these lesions in barrets, as you well know, end up being a little bit more extensive than what you initially think it to be. So this is a great example of that. Yeah, it is. What frame rate does the AI system run your images in the room? Frame rate, Mo is our research fellow. I don't know, Mo, do you know the frame rate? The speed of the AI system for the detection is about five milliseconds per image or per frame. So it's incredibly fast, obviously compared to the human eye. So that's why it's real time as you see it. It detects sometimes it certainly detects a lot faster than most of my fellows can detect the lesion. So as soon as it's within the endoscopy view, it detects the lesion. So that's the anastomosis mode. Just go down there. So you can see the sutures. So these are the surgical sutures there. You can see the black sutures there. The good thing is there is no nuisance value of the system. There is no false positive detection here. You see, a lot of time in our early phase, we noticed that the top of the gastric fold gets detected as a positive lesion because it's raised. But the beauty of this system is there's no detection of these gastric folds. There's no noise there. So you pull back slowly now. And there we can start. You can see the detection now. And it does, as Pradeep was saying, it looks a lot more extensive than the first view we got. So it is from three to six o'clock. This is pretty impressive, Pradeep. Are you able to show us some imaging? Yes. So we'll switch over to, yeah. So this is now, we were using Fuji's Elexio system. The white light itself was very bright. Now we're changing over to LCI. This is another very good modality, which enhances red. And as most of the neoplasias are red in color, you can see how the redness has been enhanced. And it almost highlights the boundaries from distance. So the beauty of LCI is you can see almost from a distance, like a bird's eye view of the scanning the esophagus, finding those red spots where the lesion is likely to be. So if you can now change over to BLI. What's the major difference between LCI and BLI? That's a very good question. So instead of going into the various wavelengths and things, the big difference is LCI enhances red and BLI enhances blue. So when you have the blue wavelength enhanced, you see the vessel patterns, superficial vessel patterns that you see here with BLI. When it enhances the redness in LCI, it gives you highlights a red spot in a slightly less red background. So the idea of LCI is a bird's eye view to attract your attention to an area of problem, which is a red hotspot. You then switch over to BLI and you assess the vessel pattern, pit pattern, and decide whether it's neoplastic or not, which is what Mo is just doing now with BLI. And you can see on the left side, the surface pattern is a bit distorted. The vessels are very compact, whereas on the right side, you have a normal. So could you show the normal pattern there on the right side? So there's hardly anything normal here. So maybe Vani, if I could ask you this, I mean, given your expertise, you know, in imaging and stuff, I mean, this is a post esophagectomy patient, no LES, free flowing reflux. How about all these changes that we may end up seeing on imaging or on AI, which are inflammatory rather than purely neoplastic in nature? Are there good ways to differentiate that, Vani? Yeah, no, that's definitely one of the challenges. I think inflammation is a powerful confounder, and especially in Barrett's with just reflux esophagitis, or especially in this setting with no LES and in a post-surgical state. You can see a lot of changes that are going to distort the pattern or create a friability. And really, I think that's important before any surveillance examination to optimize acid suppression, especially if you're going in after someone has been diagnosed with dysplasia, is to go ahead and optimize their PPI and try and take care of that before you look. With these imaging modalities, sometimes you can tell some glandular patterns that might be more obviously neoplastic, but still there's a lot of false positives in the setting of inflammation. I think the inflammation is a problem for all kinds of imaging modalities. Even if it's a problem for the pathologist, forget about the imaging modalities, and it will be a problem even for the AI until it becomes really, really clever. But in this case, having said that, Prateek, we can see the margins. I'm not sure how the resolution of transmission is, but I can see subtle difference between the end of neoplastic area and the normal mucosa. So our plan is to mark this area and do an ESD. And as you said, we'll go from three o'clock to nine o'clock and resect the entire barracks between three to nine, and then deal with the rest of the residual barracks if there's any left. So the margins, I mean, Prateek, you know, will you just mark them under BLI? How will you do it to make sure that you're getting most of the barracks neoplastic area out with ESD? Yeah, so I think you're right. BLI and magnification can be good, but lots of times when you switch on these imaging modalities, Prateek, it gets a little bit dark and you lose the context between the surroundings. So if I can see it on the white light, I prefer to mark still on the white light, but if white light doesn't help, then we would switch over to BLI and magnification. And that's where you have to be really, really focused on a very small area. If you use white light without magnification, you can see a broader context. So we prefer to do it that way. In this case, I think we just go from the squamous on the left to this normal barracks on the right, and we'll just remove the entire area in the middle. Tony, you had a question? Yeah, I mean, these have been really impressive images that you've shown us, but could you just summarize for us how you typically approach the exam? Do you start with white light without the AI on? And then do you turn on the AI? And then what do you do when you see a positive? Do you then, you know, sort of take us through sort of your approach in incorporating AI? Is it on at the beginning or do you do your exam first? That's a very, very good question. And, you know, only yesterday I had a meeting with our ESG council that we need to develop some user guidelines. AI is coming thick and fast, but the endoscopists still don't know how to use AI in a more efficient way. So I was going to summarize very quickly that I think the gold standard still remains white light endoscopy. And then from white light endoscopy and experience, if you want to move to BLI or advanced imaging or AI first and then advanced imaging, my view is white light endoscopy and AI. And then once AI shows you a hotspot, you use your advanced imaging and magnification to confirm and agree or disagree with AI. I think this is the way we will go forward with AI for the foreseeable future. Pradeep, we're going to move on. If you are available later for the Q&A, please join us and good luck with this seemingly very difficult ESD. Thank you. We'll see you soon. Transcribed by https://otter.ai

Barrett's esophagus diagnosis

neoplasia

AI system

lesion detection

advanced imaging