实时计算机内镜辅助系统对结直肠病变检出率的影响

Impact of real-time computer endoscopy-assisted system on the detection rate of colorectal lesions

摘要:

目的 评估实时计算机内镜辅助系统（内镜精灵，EndoAngel）在结肠镜检查中对结直肠病变检出的影响。方法 采用单中心自身对照研究，选取于该院消化内科与内镜中心接受EndoAngel辅助结肠镜检查和常规结肠镜检查的患者2 000例，按照检查方法不同，将患者分为人工智能（AI）组和传统肠镜组，各1 000例。以病理诊断为金标准，比较两组患者息肉检出率和腺瘤检出率，进一步按照操作医师年资进行亚组分析。结果 AI组息肉检出率为39.3%，明显高于传统肠镜组的29.0%，差异有统计学意义（χ² = 23.59，P = 0.000）。其中，AI组增生性息肉和腺瘤性息肉检出率分别为19.1%和25.2%，明显高于传统肠镜组的12.4%和20.8%，差异有统计学意义（χ² = 16.92，P = 0.000；χ² = 5.46，P = 0.019）。进一步将两组按医师年资进行亚组分析，AI低年资组息肉检出率为36.6%，高于传统肠镜低年资组的息肉检出率20.4%，差异有统计学意义（χ² = 32.20，P = 0.000）。其中，AI低年资组增生性息肉检出率为17.8%，腺瘤性息肉检出率为23.6%，明显高于传统肠镜低年资组的12.8%和13.6%，差异均有统计学意义（χ² = 4.82，P = 0.028；χ² = 16.51，P = 0.000）；高年资医师两组间腺瘤性息肉检出率差异无统计学意义（P > 0.05）。结论 EndoAngel可以提高结肠镜检查的息肉检出率，尤其针对低年资医师效果更为显著。

关键词:人工智能（AI）;结肠镜;结直肠息肉;息肉检出率;实时计算机内镜辅助系统（内镜精灵，EndoAngel）

Abstract:

Objective To evaluate the advantages of a real-time computer endoscopy-assisted system (EndoAngel) for colorectal lesions detection in colonoscopy.Methods 2 000 patients who underwent EndoAngel assisted colonoscopy and conventional colonoscopy were selected for the study in a single-center, self-controlled study. According to different examination methods, the patients were divided into artificial intelligence (AI) group and traditional colonoscopy group, each with 1 000 cases. The results were statistically analyzed and compared with the polyp detection rate and adenoma detection rate of the two groups using pathological diagnosis as the gold standard. Further subgroup analysis will be conducted based on the seniority of the operating physicians.Results AI group’s polyp detection rate was higher (39.3%) than conventional colonoscopy group polyp detection rate (29.0%), with statistically significant difference (χ² = 23.59, P = 0.000). Of these, the detection rates of hyperplastic polyps and adenomatous polyps were 19.1% and 25.2%, which were significantly higher than those of 12.4% and 20.8% in the conventional colonoscopy group, and the differences were statistically significant (χ² = 16.92, P = 0.000; χ² = 5.46, P = 0.019). Further subgroup analysis of the two groups by physician seniority, the polyp detection rate of AI low seniority group (36.6%) was higher than that of conventional colonoscopy low seniority group (20.40%), with a statistically significant difference (χ² = 32.20, P = 0.000). Among them, the detection rates of hyperplastic polyp (17.8%) and adenomatous polyp (23.6%) in AI low seniority group were higher than those in the conventional colonoscopy low seniority group (12.8% vs 13.6%), and the differences were significant (χ² = 4.82, P = 0.028; χ² = 16.51, P = 0.000). There were no significant differences in adenomatous polyp detection rates between the two groups of senior physicians.Conclusion EndoAngel assisted system can improve the polyp detection rate of colonoscopy, especially for the effect of low seniority physicians is more significant.

Key words:artificial intelligence (AI);colonoscopy;colorectal polyps;polyp detection rate;the real-time computer endoscopy-assisted system (EndoAngel)