This is ending up with a few practical tips, and I think some of it may be repetition because either during the Q&A session or during one of the lectures, you've already heard about it. So our apologies for any duplication, but I think this is also the key is as folks are trying to get started with artificial intelligence, either in their research or in the practice and what skill sets you need, we've already discussed that in a little bit of detail overall, who are your collaborators. So let's discuss that. So Sravanti, over to you and then to Mike. Okay. I think a lot of this was partly covered in different sections, so maybe I'll just consolidate it and try to finish a little early so that if you have more questions, we can specifically answer those, and I'm sure Michael Byrne has a lot of practical tips as well. So how to get started with AI and gastroenterology projects. Those are my disclosures. As the theme has been throughout this whole masterclass, first, we need to understand the basics, whether it's the basics of artificial intelligence or what kind of data you need for the specific project that you want to work at. So for that, I just put a couple of pictures here. You need to have a good understanding of what AI can do and cannot do. As it's been told multiple times, we are not ready for this generalized AI where you expect everything to run autonomously, meaning you want the scope to tell exactly everything and with 100% accuracy, we are not there yet. So what is it that we can currently achieve in the current regulatory environment and what is it that we can do within the space of gastroenterology is the second part. The question that you want to work on will be at the intersection of the need that you know within your field, whether it's electronic health records or computer vision problems or whatever AI solution you want to apply to a current problem within your space, it will be the intersection up here, and that's the question you want to move forward with. So then you need expertise from AI standpoint to help you narrow down that question to see practically what is it that is feasible and domain experts, of course, we are the domain experts who you may consult somebody else to see what else is going on in the same space. You don't want to replicate some of the work that has already been done and so forth. Now, the next step is identifying what kind of AI project it is. I know we've focused a lot on computer vision problems, meaning what is it from a pattern recognition, from an image recognition standpoint, is it colic detection, classification, barrett's, inflammatory bowel disease, dysplastic lesions, or endoscopic ultrasound to characterize the volume of a pancreatic cyst or, you know, different other things. All of those that you are looking at images and videos come under the realm of computer vision. But there is a lot of other scope within the AI space that is also rapidly evolving and several collaborations between tech chains and hospitals and hospital systems are already happening to kind of harness that information. And one other area that is ripe is risk prediction and prognostication, stratification, basically using electronic medical records. And as we've already discussed, this data can be in a structured format or an unstructured format. You might need expertise depending on maybe your question is, can I see the risk of cancer in a patient who had a colonoscopy with at least one tubular adenoma? You want to answer that question using electronic medical records. Then you need, first, data. Second, you need probably, you know, a natural language processing expert because your pathology data might not be in a readily extractable format. So you need somebody to extract it from the PDFs and so forth. And then you probably want to know where the endoscopy information is, what's the size of the polyps, so forth. And of course, you want to plug in their demographics, their medications and so forth. So when you're formulating a question, you have to think about all the variables that you need and what kind of support from your AI partners you need, along with what kind of data and how much data you can handle. And the third piece that is also moving along really well is infusing artificial intelligence technologies within digital health to improve patient outcomes. So that's a lot of data out there. If somebody is interested in those kind of questions, you need to kind of narrow down that as well. We touched on natural language processing, natural language understanding is the robots talking back to you, and all sorts of other basic research that you need to understand exactly what is it that you want to answer before you find the right expert. So again, what can AI do? You have to find the technical diligence, meaning you clearly have to formulate the question that you want to ask, and then reach out to your AI experts and find out if it is feasible or not. And then the other part is, I just put it as business diligence, not in the sense that you want to commercialize your product or idea, but basically, if you are starting to work on an area, you need to know what is it that you want to get out of that idea for sure. So you need to know, okay, at the minimum, I want a stratification model to predict who is at risk for colon cancer with a cutoff of 0.7 accuracy or something like that. That would be your question. And in some cases, as Dr. Mori already referred to earlier, if you are looking at commercialization of your product, then you have to think through all the pipelines in terms of who are your stakeholders, who are you going to sell it to, your IP, your legal, and the whole commercialization pathway as well. So that is what I would say business diligence, whether it is in an academic world or a non-academic world. So from the technical diligence part, can the AI system meet the required performance? Meaning let's say you already have a polyp detection module that is detecting it at 99% accuracy. You probably don't want to invest time in repeating the same experiment. You probably want to look at something else, which is more novel, or which probably is more relevant to your practice. And then the next question is, once you know what that question is, you need to understand how much data you need and from where you will collate that data, whether it's prospective or you want to validate already existing algorithm and so forth. And then the other thing which I learned is you also have to plug in the time that you want to finish the project. In medicine, we take quite a lot of time from start to finish of projects, in the sense you might have finished analyzing data, it's going through multiple revisions and so forth. And in some cases, even after submission to journals, it might take almost six months to a year. But in other fields, the pace is much more rapid. So then you have to plan your time and your commitment to that project and finish it along with your technical partners who will probably be working at a different pace. So something that you have to talk to them ahead of time so that you will set down your expectations right and how much time you have that you can work with them and so forth. Now, the business intelligence I briefly touched on, it's basically what is the question that you want to answer? Is it something from a GI perspective? Will it improve detection? Or will it save money for the patients or the health care in general? Or will it improve the stratification or prediction or help with population health strategies by finding patients who are at risk and getting them into the systems? Things like that. There are a lot of different ways as to how you can formulate the question, depending on your background research interests. So another area would be if you have a specific interest within, let's say, colon cancer or pancreatic cancer or something like that, then you can formulate the question around that space. That would be one way. The other one based on within AI itself, maybe you just want to create a conversational AI bot or maybe you want to work in the digital health space and so forth. So that could be one way of how you can outline a problem. And always pay attention to your pain points. A lot of times we complain about things. But the way I look at complaints are also opportunities where you can improve. If you think based on some basic understanding of artificial intelligence or how different industries and sectors are working on solving problems using AI, if you can create a conversation using AI, if you can bring it to your pain point to solve that problem, that's a good use case in itself. So always look at problems as opportunities and that will change your perspective as well. Again, this is a key part, resources for people who are just starting off. A lot of times you might have resources within your own institutions. And when I mean resources, it's not exactly somebody giving you funding or anything like that. It could be simple things like just finding data. Maybe your institution is willing to give its data, your EPIC or CERN or whatever it is. Reach out to your GI societies. They might have some resources as well. Reach out to mentors as to how they have started and whom they have collaborated with and how they have learned over a period of time and so forth. That would be one way to get started. And then as I touched upon previously and Dr. Mori also touched upon is, what is it that you want to achieve from asking this question? Is it just academic research, meaning you want to publish a paper and say that this works from a prototype standpoint? Or do you want to take it further down and maybe work on your own startup or commercialize the product? So that is another goal that you need to clarify when you're starting off because the pathways would be completely different. So basically, do your literature review. Learn what's out there, both in the AI space as well as within your domain, which could be a specific area of interest or a niche area within gastroenterology and endoscopy. Do your homework. It usually takes up to four to six weeks to kind of round this up, even after you know the basics of artificial intelligence. Once you come up with a preliminary idea, you have to go through multiple iterations to kind of understand if it's really worth doing it or not. And then create an outline listing of all the possible resources, the data sources, and what kind of commitment you need before you get started. And identify your team members. Connect with mentors or people who have worked in the similar space. Familiarize yourself with the field. Keep reading. Build your team. It'll slowly grow. And be committed. There'll be all this looks very rosy because once the study is published and now there it's neatly packaged, it seems like a very seamless work. But it is a lot of work. There'll be a lot of disappointments. So you need to be very committed as you're doing this. And definitely understand the limitations of machine learning because there's a lot of hype out there. You need to know exactly what you can do and what you cannot do so that you can be happy at the end of whatever work you do. So I will leave it up there and have Mike talk about the practical aspects of who your team is and so forth.

artificial intelligence

gastroenterology

AI projects

computer vision

risk prediction