Stanford University, Stanford, California, USA.
Trinity Health of New England, Waterbury, Connecticut, USA.
Am J Gastroenterol. 2023 Oct 1;118(10):1841-1847. doi: 10.14309/ajg.0000000000002239. Epub 2023 Mar 9.
There has been increasing interest in artificial intelligence in gastroenterology. To reduce miss rates during colonoscopy, there has been significant exploration in computer-aided detection (CADe) devices. In this study, we evaluate the use of CADe in colonoscopy in community-based, nonacademic practices.
Between September 28, 2020, and September 24, 2021, a randomized controlled trial (AI-SEE) was performed evaluating the impact of CADe on polyp detection in 4 community-based endoscopy centers in the United States Patients were block-randomized to undergoing colonoscopy with or without CADe (EndoVigilant). Primary outcomes measured were adenomas per colonoscopy and adenomas per extraction (the percentage of polyps removed that are adenomas). Secondary end points included serrated polyps per colonoscopy; nonadenomatous, nonserrated polyps per colonoscopy; adenoma and serrated polyp detection rates; and procedural time.
A total of 769 patients were enrolled (387 with CADe), with similar patient demographics between the 2 groups. There was no significant difference in adenomas per colonoscopy in the CADe and non-CADe groups (0.73 vs 0.67, P = 0.496). Although the use of CADe did not improve identification of serrated polyps per colonoscopy (0.08 vs 0.08, P = 0.965), the use of CADe increased identification of nonadenomatous, nonserrated polyps per colonoscopy (0.90 vs 0.51, P < 0.0001), resulting in detection of fewer adenomas per extraction in the CADe group. The adenoma detection rate (35.9 vs 37.2%, P = 0.774) and serrated polyp detection rate (6.5 vs 6.3%, P = 1.000) were similar in the CADe and non-CADe groups. Mean withdrawal time was longer in the CADe group compared with the non-CADe group (11.7 vs 10.7 minutes, P = 0.003). However, when no polyps were identified, there was similar mean withdrawal time (9.1 vs 8.8 minutes, P = 0.288). There were no adverse events.
The use of CADe did not result in a statistically significant difference in the number of adenomas detected. Additional studies are needed to better understand why some endoscopists derive substantial benefits from CADe and others do not. ClinicalTrials.gov number: NCT04555135.
人工智能在胃肠病学领域的兴趣日益浓厚。为了降低结肠镜检查的漏诊率，人们对计算机辅助检测（CADe）设备进行了大量探索。本研究评估了 CADe 在社区非学术实践中结肠镜检查中的应用。
2020 年 9 月 28 日至 2021 年 9 月 24 日，在美国 4 家社区内镜中心进行了一项随机对照试验（AI-SEE），评估 CADe 对息肉检测的影响。患者被随机分为接受或不接受 CADe（EndoVigilant）的结肠镜检查。主要结局指标为每例结肠镜检查的腺瘤数和每例提取的腺瘤数（切除的息肉中腺瘤的百分比）。次要终点包括每例结肠镜检查的锯齿状息肉数；每例结肠镜检查的非腺瘤性、非锯齿状息肉数；腺瘤和锯齿状息肉检出率；以及操作时间。
共纳入 769 例患者（CADe 组 387 例），两组患者的人口统计学特征相似。CADe 组和非 CADe 组的每例结肠镜检查的腺瘤数无显著差异（0.73 比 0.67，P = 0.496）。尽管 CADe 的使用并没有提高每例结肠镜检查的锯齿状息肉检出率（0.08 比 0.08，P = 0.965），但它增加了每例结肠镜检查的非腺瘤性、非锯齿状息肉的检出率（0.90 比 0.51，P < 0.0001），导致 CADe 组每例提取的腺瘤数减少。CADe 组和非 CADe 组的腺瘤检出率（35.9%比 37.2%，P = 0.774）和锯齿状息肉检出率（6.5%比 6.3%，P = 1.000）相似。CADe 组的平均退出时间长于非 CADe 组（11.7 比 10.7 分钟，P = 0.003）。然而，当没有发现息肉时，两组的平均退出时间相似（9.1 比 8.8 分钟，P = 0.288）。没有不良事件发生。
CADe 的使用并没有在检测到的腺瘤数量上产生统计学上的显著差异。需要进一步的研究来更好地理解为什么一些内镜医生从 CADe 中获得了实质性的益处，而另一些则没有。临床试验编号：NCT04555135。
