Objective: To explore the value of the artificial intelligence (AI)-assisted recognition system in the detection quality of colonoscopy. Methods: From January 2023, the data on 700 patients who underwent colonoscopy in the Digestive Endoscopy Center of the First Affiliated Hospital of Zhejiang Chinese Medical University were collected prospectively. Based on a computerized number method, patients were divided into the AI assistance group and control group. The detection rate of adenomas (ADR) and polyps, number and size of adenomas, Boston bowel preparation scale (BBPS), intubation time, withdrawal time, and cecal intubation rate were compared between groups. Normally distributed data were analyzed with the t-test for independent samples. Non-normally distributed data were analyzed with the Rank sum test. Categorical data were analyzed with the Chi-square test. Results: In total, 691 patients were included in the analysis. According to the intention to treat (ITT) analysis and per-protocol (PP) analysis, the withdrawal time of the AI group was higher than that of the control group (ITT:436 (305, 620) vs 368 (265, 510) s, Z=-4.24, P<0.001;PP:439 (306, 618) vs 364 (262, 500) s,t=-4.50, P<0.001); however, there were no significant differences in the ADR (ITT:123(35.5%) vs 111(32.2%), χ2=0.88, P=0.349;PP:108(34.2%) vs 99(31.1%), χ2=0.67, P=0.414), the number of adenomas (ITT:0(0, 1) vs 0(0, 1),Z=-1.08, P=0.282;PP:0(0, 1) vs 0(0, 1),Z=-0.87, P=0.387), the polyp detection rate (ITT:85(24.6%) vs 85(24.6%),χ2=0.001, P=0.983;PP:79(25.0%) vs 77(24.2%),χ2=0.05, P=0.818), BBPS (ITT:6.5±0.9 vs 6.5±0.7,t=-0.59, P=0.555;PP:6.7±0.6 vs 6.6±0.6,t=-1.83, P=0.068), and cecal intubation rate (ITT:346(100.0%) vs 343(99.4%), χ2=0.50, P=0.478) between these two groups. After excluding inadequate bowel preparation and failed cecal intubation cases, the AI-assisted system was found to significantly improve the detection rate of small adenomas (≤5 mm) (PP:27.8%(88/316)vs 21.1%(67/318), χ2=3.94, P=0.047). Conclusions: The application of an AI-assisted system in colonoscopy can increase the withdrawal time and improve the detection rate of small adenomas.
目的： 探讨人工智能辅助识别系统在结肠镜检查质量评价中的应用价值。 方法： 2023年1月在浙江中医药大学附属第一医院消化内镜中心，前瞻性纳入电子结肠镜检查者700例，采用计算机随机数字法分为人工智能辅助检查（AI）组和常规组，分别比较两组间腺瘤发现率（ADR）、息肉发现率、腺瘤大小和数量、肠道准备评分（BBPS）、进镜时间、退镜时间及盲肠插管率。正态分布计量数据采用两独立样本t检验进行分析，偏态分布计量数据采用秩和检验进行分析，计数数据采用χ2检验进行比较。 结果： 共有691例患者纳入分析，按意向（ITT）分析和按方案（PP）分析，AI辅助识别结肠镜组的退镜时间长于常规组［ITT：436（305，620）比 368（265，510）s，Z=-4.24，P<0.001；PP：439（306，618）比 364（262，500）s，t=-4.50，P<0.001］，但两组间ADR［ITT：35.5%（123/346）比 32.2%（111/345），χ2=0.88，P=0.349；PP：34.2%（108/316）比 31.1%（99/318），χ2=0.67，P=0.414］、腺瘤数量［ITT：0（0，1）比 0（0，1），Z=-1.08，P=0.282；PP：0（0，1）比 0（0，1），Z=-0.87，P=0.387］、息肉发现率［ITT：24.6%（85/346）比 24.6%（85/345），χ2<0.01，P=0.983；PP：25.0%（79/316）比 24.2%（77/318），χ2=0.05，P=0.818］、BBPS［ITT：（6.5±0.9）比（6.5±0.7）分，t=-0.59，P=0.555；PP：（6.7±0.6）比（6.6±0.6）分，t=-1.83，P=0.068］、盲肠插管率［ITT：100.0%（346/346）比 99.4%（343/345），χ2=0.50，P=0.478］差异无统计学意义。在排除不充分肠道准备及盲肠插管失败后，人工智能辅助系统可显著提高小腺瘤（≤5 mm）的检出率［PP：27.8%（88/316）比 21.1%（67/318），χ2=3.94，P=0.047］。 结论： 人工智能辅助系统可以增加结肠镜检查中的退镜时间，并提高小腺瘤的检出率。.