目的:系统评价基于深度学习的人工智能辅助系统对结肠镜检查质量的影响。方法:计算机检索PubMed、Em-base、Cochrane Library、Wed of Science、中国知网学术总库、万方数据库和维普数据库，获得有关人工智能辅助系统和传统结肠镜检查的随机对照研究，检索时限均为建库至2023年1月。由2位评价员独立筛选文献、提取资料并评价纳入研究的偏倚风险，使用Cochrane偏倚风险评估工具评价研究质量。应用Stata12。0及Revman5。3软件进行Meta分析，计量资料采用加权均数差(WMD)及95％CI表示，计数资料采用比值比(OR)及95％CI表示，应用Eggers检验及漏斗图评估发表偏倚。结果:最终共纳入17项随机对照研究，共涉及12 213个研究对象，其中人工智能辅助系统组6 095例，传统结肠镜检查组6 018例。Meta分析结果显示，人工智能系统辅助结肠镜检查可提高腺瘤检出率[OR=1。55，95％CI(1。42～1。70)，P=0。000]，增加息肉检出率[OR=1。63，95％CI(1。41～1。88)，P=0。000]，延长退镜时间[OR=0。17，95％CI(0。07～0。28)，P=0。001]，但在进镜时间[MD=0。00，95％CI(-0。05～0。04)，P=0。879]、腺瘤大小[MD=-0。14，95％CI(-0。31～0。03)，P=0。102]方面，其与传统结肠镜检查差异均无统计学意义。结论:基于深度学习的人工智能辅助系统可提高结肠镜检查的腺瘤检出率及息肉检出率、延长退镜时间，但与进镜时间、腺瘤大小无关。
Meta-Analysis of Effect of Artificial Intelligence Assisted System on Quality of Colonoscopy
Objective To systematically evaluate the effect of artificial intelligence(AI)assisted system based on deep learning on the quality of colonoscopy.Methods Randomized controlled studies on AI-as-sisted systems and traditional colonoscopy were searched through PubMed,Embase,Cochrane Library,Wed of Science,CNKI Academic Database,Wanfang Database and VIP Database from the establishment of the database to January 2023.Two valuators independently screened literature,extracted data and evalua-ted the risk of bias in included studies.Cochrane bias risk assessment tool was used to evaluate study qual-ity.Stata 12.0 and Revman5.3 software were used for meta-analysis.Weighted mean difference(WMD)and 95％CI were used for measurement data.Odds ratio(OR)and 95％CI were used for counting data,and Egger's test and funnel plot were used to evaluate publication bias.Results A total of 17 randomized controlled studies were included,involving a total of 12,213 subjects,including 6,095 cases in the AI-as-sisted system group and 6,018 cases in the traditional colonoscopy group.Meta-analysis results showed that AI-assisted colonoscopy could improve the detection rate of adenoma[OR=1.55,95％CI(1.42～1.70),P=0.000],increase the detection rate of polyps[OR=1.63,95％CI(1.41～1.88),P=0.000],and prolong the withdrawal time of colonoscopy[OR=0.17,95％CI(0.07～0.28),P=0.001],but in terms of the entry time of colonoscopy[MD=0.00,95％CI(-0.05～0.04),P=0.879],the size of ad-enoma[MD=-0.14,95％CI(-0.31～0.03),P=0.102],there was no significant difference between AI-assisted colonoscopy and traditional colonoscopy.Conclusion AI-assisted system based on deep learning can improve the detection rate of adenoma and polyp in colonoscopy,and prolong the withdrawal time of colonoscopy,but it has no relationship with the entry time of colonoscopy,the size of adenoma.
NETL
NSTL
万方数据
