Morimoto Shin, Tanaka Hidenori, Takehara Yudai, Yamamoto Noriko, Tanino Fumiaki, Kamigaichi Yuki, Yamashita Ken, Takigawa Hidehiko, Urabe Yuji, Kuwai Toshio, Oka Shiro
Department of Gastroenterology, Hiroshima University Hospital, Hiroshima, Japan.
Gastrointestinal Endoscopy and Medicine, Hiroshima University Hospital, Hiroshima, Japan.
J Anus Rectum Colon. 2025 Jan 25;9(1):127-133. doi: 10.23922/jarc.2024-055. eCollection 2025.
Studies have suggested that computer-aided polyp detection using artificial intelligence improves adenoma identification during colonoscopy. However, its real-world effectiveness remains unclear. Therefore, this study evaluated the usefulness of computer-aided detection during regular surveillance colonoscopy.
Consecutive patients who underwent surveillance colonoscopy with computer-aided detection between January and March 2023 and had undergone colonoscopy at least twice during the past 3 years were recruited. The clinicopathological findings of lesions identified using computer-aided detection were evaluated. The detection ability was sub-analyzed based on the expertise of the endoscopist and the presence of diminutive adenomas (size ≤5 mm).
A total of 78 patients were included. Computer-aided detection identified 46 adenomas in 28 patients; however, no carcinomas were identified. The mean withdrawal time was 824 ± 353 s, and the mean tumor diameter was 3.3 mm (range, 2-8 mm). The most common gross type was 0-Is (70%), followed by 0-Isp (17%) and 0-IIa (13%). The most common tumor locations were the ascending colon and sigmoid colon (28%), followed by the transverse colon (26%), cecum (7%), descending colon (7%), and rectum (4%). Overall, 34.1% and 38.2% of patients with untreated diminutive adenomas and those with no adenomas, respectively, had newly detected adenomas. Endoscopist expertise did not affect the results.
Computer-aided detection may help identify adenomas during surveillance colonoscopy for patients with untreated diminutive adenomas and those with a history of endoscopic resection.
研究表明，使用人工智能的计算机辅助息肉检测可提高结肠镜检查期间腺瘤的识别率。然而，其在现实世界中的有效性仍不明确。因此，本研究评估了在定期监测结肠镜检查中计算机辅助检测的实用性。
招募了2023年1月至3月期间接受计算机辅助检测的监测结肠镜检查且在过去3年中至少接受过两次结肠镜检查的连续患者。对使用计算机辅助检测识别出的病变的临床病理结果进行评估。根据内镜医师的专业知识和微小腺瘤（大小≤5mm）的存在情况对检测能力进行亚分析。
共纳入78例患者。计算机辅助检测在28例患者中识别出46个腺瘤；然而，未识别出癌。平均退镜时间为824±353秒，平均肿瘤直径为3.3mm（范围2-8mm）。最常见的大体类型是0-Is（70%），其次是0-Isp（17%）和0-IIa（13%）。最常见的肿瘤部位是升结肠和乙状结肠（28%），其次是横结肠（26%）、盲肠（7%）、降结肠（7%）和直肠（4%）。总体而言，未治疗的微小腺瘤患者和无腺瘤患者中分别有34.1%和38.2%新检测出腺瘤。内镜医师的专业知识不影响结果。
计算机辅助检测可能有助于在监测结肠镜检查期间识别未治疗的微小腺瘤患者和有内镜切除史患者中的腺瘤。
