Time for some questions, and before we get to the questions, let me just introduce the remaining panelists. Dr. Yutaka Saito is the director of endoscopy division at the National Cancer Center Hospital in Tokyo, Japan. He is an associate professor of gastroenterology at Tokyo Medical University. He has a wide spectrum of research exposure and interest, including colon capsule endoscopy and artificial intelligence diagnosis for colorectal neoplasia. We also have Dr. Lucy Liu Wang, who is a postdoctoral investigator at the Allen Institute for Artificial Intelligence in the Semantic Scholar Research Group. Dr. Wang's work is in the areas of biomedical ontologies, bio-natural language processing, scientific natural language processing, open access, and meta-science. And last but not least, we have Dr. Cesare Hassan, who is a gastroenterologist at the Nuovo Regina Margarita Hospital in Rome, Italy. His clinical and research are focused mainly on the development of new approaches for colorectal cancer screening, mainly focusing on clinical efficacy, cost effectiveness, and decision making. Welcome to all the panelists. So perhaps we'll start, Seth, with a question to you. Very nice presentation on how system hardwares and softwares integrate into the endoscopy suite. In your experience, have you experienced any ergonomic challenges within the procedure rooms on working with ancillary staff? So I think that the biggest challenge, and we experienced this a few years ago, for any of you that used the colonoscopy system that had the three monitors, you know, my preference, hopefully, in the future, is that everything's on one monitor, because it is a little bit distracting looking to, in my setup, in my unit, I have to look to my right. I do use my nurse or my tech who I'm working with, you know, to help me out as well. But I do find it a bit distracting from an ergonomics point of view. I think the future would be that the AI software is just on one monitor. And it highlights us and puts a box around areas that we need to make clinical decisions on real time. Yes, Cesare. I absolutely agree with the set, only one monitor, and not necessarily only one software. Maybe when I was hearing, Seth, your presentation, I was thinking, why not to have two plug-in in the same system? Maybe you can have an European based software, and then you can have a Japanese software created by Yutaka, and they may compensate each other. My fear is that the audience feel that all the software are the same. It's not true. Software are completely different, and we need to adapt to it. So this is my suggestion, plug in more software. Excellent, excellent. And Dr. Rivlin, very nice presentation as well. Can you speak to the importance of how a system, when it comes to measuring detection over a segment, the risk of a system that would not differentiate between the segments, and how that would computate towards a detection measurement? Can you repeat it? You're talking about coverage? That's correct. Oh, sure. And the question is, I mean, would it influence your performance? Where you are in the colon, why this segment is critical? I think that that came from a visit to a couple of hospitals, when they mentioned that the main thing that they are afraid of is missing coverage in a way. Not the many polyps that we can find a lot of, and I think that we should relate, probably you should relate to all, but the big one that you miss just because you had some noise, or you turn into the right, or you saw something interesting and you run forward, and then you miss this one. And I think this thing is critical, just to get the fact that you need, if you want, to go back and coverage some area. Or looking back, I mean, you covered whatever, 70% of the colon. May I ask, you, Cesare, you talk a different location in the world. What would be the largest benefit of AI in endoscopy? Is there a difference in Europe, in Asia, or in the States? Are the costs, the quality, minimizing complications? What do you think? So, here in Europe, Helmut, we suffer from a large variability in the performance of diagnostic endoscopy. The ADR in a screening program with the same patient can range from 20 to 70%. And this is still today, after 10 years of quality improvement. So, we need some more objective standards. So, Helmut, I guess that now in Europe, the main challenge is the standardization of the performance of the operator. Okay, thanks. Okay, would you agree? What is your opinion? Can I say something? Yes, good night, everyone. So, I think the AI, the most important thing is for the detection, maybe to find the flat or depressed region. Because for polyploid region, you can find very easily. But the flat or depressed type region is really difficult to detect even by an expert. So, that is the AI's benefit. If we, the AI learn lots of flat region, that AI system could detect the flat and depressed region. And in addition, I also think the blind spot detection is really important. Even using AI, if we miss the behind the fold, so the AI cannot detect the region behind the fold. So, the professor, now, in the gastric screening AI, there are some reports from China, a professor from Liu, you, AI detect the blind spot and teach the blind spot for the endoscopies. That is also really important in the correct AI system. That is my opinion. Okay. So, I agree. I think that one is, what's the question we're trying to answer? Is the reason we're missing lesions because they're right in front of us and we don't recognize them? I think artificial intelligence has an important role there. But if it's a surface area exposure issue, until artificial intelligence could also tell us to look behind that fold, we're not going to fully benefit from artificial intelligence. I think the other area where it potentially helps us would be where it impacts the whole procedure, including the procedure report, because I'm sure that just like in the United States, there's tremendous variability in endoscopist procedure reports. And it would be really nice, even though we do have a common language, we could certainly do a better job standardizing our reports and the artificial intelligence programs potentially could help us do that. Okay. It's interesting you bring that up. It's interesting you bring that up, Seth, because, you know, in your presentation, you commented on how the intelligence systems can interpret not only what type of lesion we're seeing or how we're removing the lesion, but also what type of accessory we're using in order to do that. Perhaps that type of information, as you're describing, could be delivered and generate a report. And certainly there's advantages to that in that you can reduce the time to take to document a report or reduce potential costs and spend more time with patients. It looks like we have a couple questions from the audience. The first is, could Dr. Wang comment on CORD-19 and artificial intelligence-based literature search? Hi. I think maybe this is a good time to jump in when talking about endoscopy reports. So it's great to see that these computer vision systems have been put in place to help with endoscopic and colonoscopy diagnosis. But I think there's also this other part, which is the language component or the report component. I think there's a real potential for natural language processing, natural language understanding techniques to help with diagnostics and treatment as well. So at the Allen Institute of AI, we primarily work on tools to support literature discovery, identifying evidence for the latest treatment developments and things like that. I guess the question was regarding CORD-19, which is a data set of COVID-19 related literature that we released earlier in this year to support literature discovery and text mining. And I think maybe the thing that I can say here is that literature is being published at a really high rate and it's challenging to keep up as a clinical researcher or as a practicing clinician. And that there are many systems for summarization or for automating parts of the literature review process that could also benefit maybe practicing clinicians in this domain. Excellent. Another question from the audience, is the FDA looking more closely at CDS tools in general and will they be subject to more rigorous vetting processes by CDS computer detection systems, of course? Perhaps, Dr. Rivlin, you want to take a shot at that? I think we have a chance to hear better in a second when we get the FDA on board. Yes. But I think that probably Helmut can answer this from the ESG, right? I think we saw what's going on in Europe and I think that at least, but I think that you'll hear it soon, but I think the direction is positive. That's my feeling. Very good. A question from a computer scientist, perhaps, Seth, you can help us with this one. For the generalizability test, I am planning to train the AI system on a colonoscopy data site and test it with wireless capsule endoscopy. Would you think it would be of clinical relevance? It seems like you're mixing apples and oranges. You know, what we need for capsule endoscopy is different than what we need for colonoscopy versus what we need for upper endoscopy. Go ahead, Cesare. Yeah, I fully confirm. At this time, AI can do only very narrow tasks. And I want to go back to what Yutaka was saying. For instance, in Europe, our database contains no more than two or three non-granular LST. So, probably we need some help from Japan to have database more in reach of this difficult lesion because an AI not trained with this lesion will not detect this lesion. So, I agree with Seth. We cannot mix and we cannot expect even between colonoscopy and colon capsule, I'm not expecting any synergy. So, very narrow task and please look at the training of the system. Very good. Excellent. Excellent. And, you know, I would agree with that completely. Another question here from the audience, Dr. Seito, is around third space endoscopy. Specifically, can artificial intelligence help us with ESD? Uh-huh. Yes. So, it will be the future option of AI. So, still we are now accumulating the data with such kind of system. But AI support the ESD or EMR, it's really more challenging than AI detection or characterization. But of course, I believe in the near future, AI will assist the ESD or especially for robotic ESD. It is much more easier for the robotic ESD assist by the AI technology. Okay. Thank you. And I think we're running short on time here. But it appears we are out of time. We are running out of time. No, I hope you guys aren't getting that bad impression here. So, Mike's given me this dirty job, right? I mean, he's probably sitting smiling there with his camera off. And so, you know, all good things have to come to an end as we need to do with this panel discussion. And Shyam and Helmut, you know, masterful job at the moderation. And we could go on and on with this excellent panel. So, I just wanted to thank both of you. And then Ehud, Yutaka, Lucy Lu, Cesare, Seth, you know, for the panel discussion. So, again, a wonderful time, guys. We'll take a 10-minute short break. And we will be back at 12.45 Central Time for, you know, the final but most important session. And as Ehud and Shyam mentioned, some of the FDA questions will definitely be coming up there. So, keep on tuning in to this and we'll see you in a few minutes. Thank you.

artificial intelligence

endoscopy

ergonomic challenges

system coverage

blind spots