Medical Department, Sahlgrenska University Hospital, Goteborg, Sweden.
Department of Molecular and Clinical Medicine, Institute of Medicine, University of Gothenburg, Goteborg, Sweden.
BMJ Open Gastroenterol. 2024 Jan 30;11(1):e001247. doi: 10.1136/bmjgast-2023-001247.
Colorectal cancer (CRC) has a significant role in cancer-related mortality. Colonoscopy, combined with adenoma removal, has proven effective in reducing CRC incidence. However, suboptimal colonoscopy quality often leads to missed polyps. The impact of artificial intelligence (AI) on adenoma and polyp detection rate (ADR, PDR) is yet to be established.
We conducted a randomised controlled trial at Sahlgrenska University Hospital in Sweden. Patients underwent colonoscopy with or without the assistance of AI (AI-C or conventional colonoscopy (CC)). Examinations were performed with two different AI systems, that is, Fujifilm CADEye and Medtronic GI Genius. The primary outcome was ADR.
Among 286 patients, 240 underwent analysis (average age: 66 years). The ADR was 42% for all patients, and no significant difference emerged between AI-C and CC groups (41% vs 43%). The overall PDR was 61%, with a trend towards higher PDR in the AI-C group. Subgroup analysis revealed higher detection rates for sessile serrated lesions (SSL) with AI assistance (AI-C 22%, CC 11%, p=0.004). No difference was noticed in the detection of polyps or adenomas per colonoscopy. Examinations were most often performed by experienced endoscopists, 78% (n=86 AI-C, 100 CC).
Amidst the ongoing AI integration, ADR did not improve with AI. Particularly noteworthy is the enhanced detection rates for SSL by AI assistance, especially since they pose a risk for postcolonoscopy CRC. The integration of AI into standard colonoscopy practice warrants further investigation and the development of improved software might be necessary before enforcing its mandatory implementation.
NCT05178095.
结直肠癌（CRC）在癌症相关死亡率中占重要地位。结肠镜检查结合腺瘤切除术已被证明可有效降低 CRC 的发病率。然而，结肠镜检查质量欠佳常常导致腺瘤漏诊。人工智能（AI）对腺瘤和息肉检出率（ADR、PDR）的影响尚未确定。
我们在瑞典萨尔格伦斯卡大学医院进行了一项随机对照试验。患者接受结肠镜检查，检查时有无 AI 辅助（AI-C 或常规结肠镜检查（CC））。检查使用两种不同的 AI 系统进行，即富士胶片 CADEye 和美敦力 GI Genius。主要结局为 ADR。
在 286 名患者中，240 名患者接受了分析（平均年龄：66 岁）。所有患者的 ADR 为 42%，AI-C 组和 CC 组之间无显著差异（41% vs 43%）。总体 PDR 为 61%，AI-C 组有更高的 PDR 趋势。亚组分析显示 AI 辅助检测时锯齿状息肉（SSL）的检出率更高（AI-C 组 22%，CC 组 11%，p=0.004）。结肠镜检查中每例患者的息肉或腺瘤检出数量无差异。检查大多由经验丰富的内镜医生进行，78%（n=86 AI-C，100 CC）。
在人工智能整合的背景下，ADR 并未因 AI 而提高。值得注意的是，AI 辅助检测 SSL 的检出率显著提高，特别是因为它们存在结直肠 CRC 后发病的风险。将 AI 纳入标准结肠镜检查实践需要进一步研究，可能需要开发改进的软件，然后才能强制实施。
NCT05178095。
