摘要
目的系统评价基于深度学习的智能辅助内镜诊断系统(intelligence-assisted endoscopic diagnosis system based on deep learning, DL-IEDS)对上消化道早癌的诊断价值。方法系统检索Pubmed、Embase、Web of Science、Cochrane Library、Sinomed、CNKI、维普及万方等中英文数据库中有关运用DL-IEDS诊断上消化道早癌的诊断性试验。纳入的研究按诊断准确性研究质量评价工具-2进行文献质量评价, 并采用Rev Man 5.3、Meta-Disc 1.4和Stata 15.1统计软件综合对数据进行Meta分析。结果最终纳入8篇文献, 共9 675张图片, 其中早癌图片2 748张。Meta分析结果显示:DL-IEDS诊断上消化道早癌的合并灵敏度、特异度、阳性似然比、阴性似然比及综合诊断比值比分别为0.920、0.874、6.824、0.103及71.109, 综合受试者工作特征曲线的曲线下面积(area under the curve, AUC)为0.958 7;其中5篇文献报道了DL-IEDS诊断早期胃癌的结果, 数据合并分析后结果显示, DL-IEDS的合并敏感度和特异度分别为0.840和0.845, AUC为0.919;4篇文献报道了内镜专家及内镜经验不足者诊断上消化道早癌的结果, 前者合并敏感度、特异度及AUC分别为0.693、0.892及0.892 3, 后者分别为0.586、0.860及0.754 5。对DL-IEDS、内镜专家及内镜经验不足者诊断上消化道早癌的AUC进行比较发现, DL-IEDS与内镜专家间差异无统计学意义(Z=1.510, P=0.131), DL-IEDS与经验不足者间差异有统计学意义(Z=6.841, P<0.001)。结论 DL-IEDS对上消化道早癌具有较高的诊断准确性, 能够明显提高内镜经验不足者对上消化道早癌的诊断能力。
Objective To systematically evaluate the intelligence-assisted endoscopic diagnosis system based on deep learning(DL-IEDS)for early cancer of the upper digestive tract.Methods Literature on the value of DL-IEDS for diagnosis of early cancer of the upper digestive tract was searched in English(PubMed,Embase,Web of Science and Cochrane Library)and Chinese databases(Sinomed,CNKI,Wanfang and VIP).The quality of literatures was evaluated according to Quality Assessment of Diagnostic Accuracy Studies-2.The Rev Man 5.3,Meta-Disc 1.4 and Stata 15.1 were used for the meta-analysis.Results Eight studies were included with a total of 9675 images(including 2748 images of early cancer).Meta-analysis results showed that the pooled sensitivity,specificity,positive likelihood ratio,negative likelihood ratio and comprehensive diagnostic ratio of DL-IEDS in the diagnosis of early cancer of the upper digestive tract were 0.920,0.874,6.824,0.103 and 71.109,respectively.The area under the curve(AUC)of summary receiver operating characteristics was 0.9587.Five studies reported the results of DL-IEDS in the diagnosis of early gastric cancer,and the combined analysis showed that the pooled sensitivity and specificity were 0.840 and 0.845 respectively,and the AUC was 0.919.Four studies reported the accuracy rate of endoscopic experts and endoscopic novices in diagnosing early upper gastrointestinal cancer,and results showed that the pooled sensitivity,specificity and AUC were 0.693,0.892 and 0.8923,and 0.586,0.860 and 0.7545,respectively.Compared with endoscopy experts,the AUC of DL-IEDS in diagnosis of early upper gastrointestinal cancer showed no statistically significant difference(Z=1.510,P=0.131),while compared with endoscopy novices,the difference was statistically significant(Z=6.841,P<0.001).Conclusion The DL-IEDS has high diagnostic accuracy for early upper digestive tract cancer,and can significantly improve the diagnostic ability of endoscopy novices.
作者
韩伟
秦小金
魏延
周金池
张哲
赵曙光
Han Wei;Qin Xiaojin;Wei Yan;Zhou Jinchi;Zhang Zhe;Zhao Shuguang(Department of Gastroenterology,Tangdu Hospital,Air Force Military Medical University,Xi′an 710038,China)
出处
《中华消化内镜杂志》
CSCD
2021年第10期828-835,共8页
Chinese Journal of Digestive Endoscopy
关键词
人工智能
深度学习
卷积神经网络
上消化道早癌
META分析
Artificial intelligence
Deep learning
Convolutional neural network
Early upper gastrointestinal cancer
Meta-analysis