So now, let me introduce Dr. Mori. Dr. Mori is an associate professor at Showa University in Yokohama, Japan. He also recently joined the University of Oslo to establish a more robust research program in the field of AI. He has a strong interest in brain research in the AI field for coronoscopy. He has published a lot of papers, more than 100 articles, and has obtained about 10 public research funding awards. He has contributed to the position of a regulatory advisor for different AI tools in Japan. So welcome, Dr. Mori, and we look forward to your presentation. It is my greatest honor to present the latest trend in computer-aided diagnosis for colonoscopy in the second AI Summit endorsed by ASGE. So this is my COI, and this slide shows the roadmap for implementation of AI in colonoscopy. This is from my personal perspective, but this roadmap contains five phases, product development, feasibility study, clinical trials, regulatory approval, and reimbursement. Today, I'd like to do a very quick review on clinical trials, and I'd like to share more time on regulatory approval issues and the possibility to have a reimbursement from public health insurance bodies in the field of AI in colonoscopy. So this slide shows the two major roles of AI in colonoscopy. The first role would be automatic polyp detection, which is called CDE. The other one would be automatic polyp characterization, or CDX. So first, CDE. As you can see, there have been already six really nice RCTs published in scientific journals, and in each RCT, patients were allocated to either colonoscopy or colonoscopy with the support of AI. And there have been already two meta-analyses to clarify the value of AI in these six RCTs. And according to this analysis, ADR was expected to increase from 25% to 37%, which is a huge amount of value, I guess. So let's quickly move on to the automatic polyp characterization. Now, as you can see in this field, six prospective studies have already been published as well. However, if you look at the quality of the study, it is completely different from that of the CDE, because there is no blood center study, no RCT, all the study but one included less than 100 patients, small studies. Therefore, we should wait for a year, more than years, to obtain a more convincing evidence in the field of CDX. It's my perspective. However, if you pick up the largest study in this field, which was conducted by our research team, this study included over 700 patients in a real-time fashion. And the primary endpoint was NPV, or negative predictive value, for rectissimilar adenomas, which was over 95%. So this result was very encouraging to use of the optical biopsy or optical diagnosis during colonoscopy. However, the drawback of this study is the use of the special endoscope instead of using the normal endoscope. So the generalizability of the study is limited. Anyway, the society is now moving forward to launch a new guideline. This is the latest guideline from the European Society of GI Endoscopy. And according to this guideline, use of AI in colonoscopy is weakly recommended based on low-quality evidence. During the CDX procedure or a CDE procedure, it's really encouraging because this kind of guideline can be beneficial for manufacturers, researchers, and possibly users or patients. So let's move on to the regulatory issues. In this slide, you can see the updated products which are on the market in the EU area and Japan. And as you can see, there are five products in the CDE area and three products in CDX area. So why only the EU and Japan's data were presented and no device in the United States is approved by FDA in terms of the AI in colonoscopy? This is why, because of the assessment system in the EU and Japan, actually when it comes to the medical device software, it is allowed to do a retrospective analysis to get a regulatory approval in the EU area or in Japan. Of course, doing a prospective study is preferred in terms of evidence or in terms of the regulatory perspective. However, it would take a lot of money and costs. Therefore, this kind of recommendation would be really appreciated to accelerate the speedy regulatory approval process. And the other problem in the regulatory approval or the products on the market is how to compare the performance of already available items. If you look at the five products in the CDE for colonoscopy, these products have almost same functionality, namely picking up the location of the polyps. However, we cannot objectively assess the performance of these devices because of the difference of the test data. To overcome this issue, several research teams are launching their own publicly accessible colonoscopy data database. However, the weak point of this database is the limited number of the polyps or images included in the databases. So to overcome this challenge, we are just establishing a new database called Sun Database. Sun Database has much more number of polyps, much more number of images with a lot of the annotation, including the location of the polyp and the nature of the polyps. And this is the structure of the database. This is really huge, including over 1.5 million frames with a full annotation, full location of the polyps. And also the quality of the database was assured rigorously. We have included three research assistants plus two expert endoscopies for construction of the database. And these procedures are completely audited by the external reviewers. And the strengths of the database would be the regulatory compatible aspects, because this database was used for a study to obtain regulatory approval of endocrine I in Japan. So hopefully, this kind of database can be contributing to some extent to the regulatory process in Japan and other countries in some future. That's why we are distributing this database for free, regardless of the commercializing bodies or the personal researchers. So let's move on to the last topic, namely reimbursement. So far, to the best of my knowledge, there is no AI-assisted colonoscopy device that secures the reimbursement from the public health insurance bodies. But if you look at the other organ, there is. This is the news last September. And this product produced by Viz AI was granted some payment from Medicare, namely reimbursement from Medicare. It's really surprising news and encouraging news for researchers in this field. I think this kind of thing can be happening in the field of colonoscopy, because we have already have a lot of data in terms of the RCT, in terms of the cost-effectiveness, as I will show later. So apparently, cost-effectiveness is the key to obtain reimbursement. And this is a schema to elucidate how we can calculate the cost-effectiveness of AI in colonoscopy. So apparently, CADE can increase the poly-detection rate, which means more number of polytectomies, more number of adenomas, more number of the pathologies, more number of the surveillance colonoscopies will be required if we use CADE, which is, I guess, the disadvantage of using CADE in terms of the financial aspect. However, this kind of increment can be balanced with the benefits of cancer prevention, thanks to the increment of ADR. And also, the cost of colonoscopy can be reduced with use of CDX, because CDX can contribute to saving of polydectomy or pathological costs. So this is a big picture of how to calculate cost-effectiveness in AI for colonoscopy. Actually, we have recently published the data with regard to CDX and cost-saving. According to this study, roughly $85 million per year can be saved in the United States if you use the CDX in routine colonoscopy. And also, the use of CDX is very beneficial psychologically to accelerate the optical biopsy. This is a really nice study published from Dr. Bergin's research team, and according to this survey study, endoscopists have a comfortable feeling of doing the, leaving the hyperplastic porous during the endoscopy if they use the AI. So this is really interesting data that we have. And also, this is the latest micro-stimulation study that we are working on, together with Professor Hassan Araya. And in this micro-stimulation study to establish the cost-effectiveness of AI in colonoscopy, we did a lot of assumptions. Namely, with use of CDE, we can get an increment of ADR and PDR. Also, we can get some saving with use of CDX. And after doing the modeling, we have got an 11% reduction in terms of CRC incidence and 7% reduction in terms of CRC death. And let me show the cost-effectiveness. And if you use only CDE, ICER, or incremental cost-effectiveness ratio, would be $92,000 per QALY, which is within the predefined willingness to pay. So from this perspective, use of CDE in colonoscopy can be cost-beneficial. And also, if you add a CDX, ICER will decrease from $92,000 to $86,000. So it's much more cost-effective. So ladies and gentlemen, these are my take-home messages. First of all, at least in EU area and Japan, multiple AI tools for colonoscopy are already on the market. And I guess our next challenge would be getting a reimbursement from public insurance bodies. So far, we are trying to negotiate with the Japanese public insurance body by providing this kind of data to get a reimbursement in some future. Thank you very much for your kind attention.

artificial intelligence

colonoscopy

implementation

automatic polyp detection

automatic polyp characterization