Scholarlink

<p>The following article provides an overview of artificial intelligence (AI) applications in gastrointestinal endoscopy. It presents AI-supported polyp detection systems, which, according to one of the latest meta-analyses of over 40&#xa0;randomized studies, increase the adenoma detection rate by an average of 7.5%. However, the gain mainly affects small adenomas without a&#xa0;significant effect on the number of advanced adenomas identified. A&#xa0;microsimulation model based on these data predicts that for every 10,000 people screened, colorectal cancer will only decrease from 82&#xa0;to&#xa0;71 and cancer-related deaths from 15&#xa0;to&#xa0;13 within 10&#xa0;years. In addition to the positive effects, risks arising from human–machine interaction, including overreliance and de-skilling, are highlighted in the article. AI-supported optical diagnosis is presented as a&#xa0;second important domain. It is designed to distinguish between neoplastic and non-neoplastic lesions in real time and enable strategies such as “resect-and-discard” or “diagnose-and-leave.” The threshold of a&#xa0;negative predictive value of ≥ 90% required by the American Society for Gastrointestinal Endoscopy (ASGE) in its Preservation and Incorporation of Valuable Endoscopic Innovations(PIVI) statement is often not met in current studies—mainly due to low specificity and lower performance of the systems in the proximal colon. Overall, the cost effects are small, and implementation is primarily realistic for diminutive polyps. Other AI applications in gastroscopy, capsule endoscopy, cholangioscopy, and endosonography show promising accuracy and greatly reduced evaluation times, but their implementation outside Asia is limited. Overall, the long-term benefits for patients remain unclear.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Risiko und Nutzen von künstlicher Intelligenz in der luminalen Endoskopie

  • 期刊论文

  • 发表日期：

    2025年12月23日

查看全文
  • Alexander Hann,
  • Riccardo Vasapolli

摘要

The following article provides an overview of artificial intelligence (AI) applications in gastrointestinal endoscopy. It presents AI-supported polyp detection systems, which, according to one of the latest meta-analyses of over 40 randomized studies, increase the adenoma detection rate by an average of 7.5%. However, the gain mainly affects small adenomas without a significant effect on the number of advanced adenomas identified. A microsimulation model based on these data predicts that for every 10,000 people screened, colorectal cancer will only decrease from 82 to 71 and cancer-related deaths from 15 to 13 within 10 years. In addition to the positive effects, risks arising from human–machine interaction, including overreliance and de-skilling, are highlighted in the article. AI-supported optical diagnosis is presented as a second important domain. It is designed to distinguish between neoplastic and non-neoplastic lesions in real time and enable strategies such as “resect-and-discard” or “diagnose-and-leave.” The threshold of a negative predictive value of ≥ 90% required by the American Society for Gastrointestinal Endoscopy (ASGE) in its Preservation and Incorporation of Valuable Endoscopic Innovations(PIVI) statement is often not met in current studies—mainly due to low specificity and lower performance of the systems in the proximal colon. Overall, the cost effects are small, and implementation is primarily realistic for diminutive polyps. Other AI applications in gastroscopy, capsule endoscopy, cholangioscopy, and endosonography show promising accuracy and greatly reduced evaluation times, but their implementation outside Asia is limited. Overall, the long-term benefits for patients remain unclear.