Makar Jonathan, Abdelmalak Jonathan, Con Danny, Hafeez Bilal, Garg Mayur
Department of Medicine, The University of Melbourne, Melbourne, Victoria, Australia.
Department of Gastroenterology, Austin Hospital, Heidelberg, Victoria, Australia; Department of Gastroenterology, Alfred Hospital, Melbourne, Victoria, Australia; Central Clinical School, Monash University, Melbourne, Victoria, Australia.
Gastrointest Endosc. 2025 Jan;101(1):68-81.e8. doi: 10.1016/j.gie.2024.08.033. Epub 2024 Aug 30.
Artificial intelligence (AI) is increasingly used to improve adenoma detection during colonoscopy. This meta-analysis aimed to provide an updated evaluation of computer-aided detection (CADe) systems and their impact on key colonoscopy quality indicators.
We searched the EMBASE, PubMed, and MEDLINE databases from inception until February 15, 2024, for randomized control trials (RCTs) comparing the performance of CADe systems with routine unassisted colonoscopy in the detection of colorectal adenomas.
Twenty-eight RCTs were selected for inclusion involving 23,861 participants. Random-effects meta-analysis demonstrated a 20% increase in adenoma detection rate (risk ratio [RR], 1.20; 95% confidence interval [CI], 1.14-1.27; P < .01) and 55% decrease in adenoma miss rate (RR, 0.45; 95% CI, 0.37-0.54; P < .01) with AI-assisted colonoscopy. Subgroup analyses involving only expert endoscopists demonstrated a similar effect size (RR, 1.19; 95% CI, 1.11-1.27; P < .001), with similar findings seen in analysis of differing CADe systems and healthcare settings. CADe use also significantly increased adenomas per colonoscopy (weighted mean difference, 0.21; 95% CI, 0.14-0.29; P < .01), primarily because of increased diminutive lesion detection, with no significant difference seen in detection of advanced adenomas. Sessile serrated lesion detection (RR, 1.10; 95% CI, 0.93-1.30; P = .27) and miss rates (RR, 0.44; 95% CI, 0.16-1.19; P = .11) were similar. There was an average 0.15-minute prolongation of withdrawal time with AI-assisted colonoscopy (weighted mean difference, 0.15; 95% CI, 0.04-0.25; P = .01) and a 39% increase in the rate of non-neoplastic resection (RR, 1.39; 95% CI, 1.23-1.57; P < .001).
AI-assisted colonoscopy significantly improved adenoma detection but not sessile serrated lesion detection irrespective of endoscopist experience, system type, or healthcare setting.
人工智能（AI）在结肠镜检查中用于提高腺瘤检出率的应用日益广泛。本荟萃分析旨在对计算机辅助检测（CADe）系统及其对结肠镜检查关键质量指标的影响进行更新评估。
我们检索了EMBASE、PubMed和MEDLINE数据库，检索时间从数据库建立至2024年2月15日，查找比较CADe系统与常规非辅助结肠镜检查在结直肠腺瘤检测中表现的随机对照试验（RCT）。
共纳入28项RCT，涉及23861名参与者。随机效应荟萃分析表明，人工智能辅助结肠镜检查使腺瘤检出率提高了20%（风险比[RR]，1.20；95%置信区间[CI]，1.14 - 1.27；P <.01），腺瘤漏诊率降低了55%（RR，0.45；95% CI，0.37 - 0.54；P <.01）。仅涉及专家内镜医师的亚组分析显示了相似的效应大小（RR，1.19；95% CI，1.11 - 1.27；P <.001），在对不同CADe系统和医疗环境的分析中也有类似发现。使用CADe还显著增加了每次结肠镜检查发现的腺瘤数量（加权平均差，0.21；95% CI，0.14 - 0.29；P <.01），这主要是由于微小病变检出率增加，而在高级别腺瘤的检测中未发现显著差异。无蒂锯齿状病变的检出率（RR，1.10；95% CI，0.93 - 1.30；P =.27）和漏诊率（RR，0.44；95% CI，0.16 - 1.19；P =.11）相似。人工智能辅助结肠镜检查使退镜时间平均延长0.15分钟（加权平均差，0.15；95% CI，0.04 - 0.25；P =.01），非肿瘤性切除率增加了39%（RR，1.39；95% CI，1.23 - 1.57；P <.001）。
无论内镜医师经验、系统类型或医疗环境如何，人工智能辅助结肠镜检查均显著提高了腺瘤检出率，但未提高无蒂锯齿状病变的检出率。
