基于CT的机器学习模型鉴别诊断肾肿瘤良恶性的Meta分析  

Accuracy of CT-based machine learning model for differentiating undistinguishable benign and malignant renal tumor:a Meta analysis

作　　者：地里亚尔·地里夏提 鲁剑德 木拉提·热夏提[1] 热衣汉·西里甫[2] 拜合提亚·阿扎提[1] DILIYAER Dilixiati;LU Jiande;MULATI Rexiati;REYIHAN Xilipu;BAIHETIYA Azhati(Department of Urology,The First Affiliated Hospital of Xinjiang Medical University,Urumqi 830054,China;Department of Nephrology,The First Affiliated Hospital of Xinjiang Medical University,Urumqi 830054,China)

机构地区：[1]新疆医科大学第一附属医院泌尿中心,新疆乌鲁木齐830054 [2]新疆医科大学第一附属医院肾病一科,新疆乌鲁木齐830054

出　　处：《现代泌尿外科杂志》2022年第12期1036-1041,1045,共7页Journal of Modern Urology

基　　金：吴阶平医学基金会临床科研专项资助基金(No.320.6750.19094-43)。

摘　　要：目的采用Meta分析评价基于CT的机器学习模型(ML)在鉴别诊断难辨别肾良性肿瘤与肾细胞癌的价值。方法检索PubMed、The Cochrane Library、Web of Science、Medline、CNKI、万方数据库自建库至2022年3月发表的有关基于CT的ML模型鉴别诊断难辨别的肾良性肿瘤(肾嗜酸细胞瘤、肾乏脂肪血管平滑肌脂肪瘤)与肾细胞癌的中英文文献。采用Stata 14.0、RevMan 5.4、meta-Disc 1.4软件进行Meta分析,计算纳入文献的合并敏感性、合并特异性、阳性似然比、阴性似然比和诊断价值比,绘制总受试者工作曲线(SROC),计算曲线下面积(AUC)。将测试集数量、模型验证策略、学习模型种类进行亚组分析,采用Meta回归分析非阈值效应引起的异质性。结果共纳入12项文献,合并敏感性、特异性、阳性似然比、阴性似然比、诊断比值比分别为0.76(95%CI:0.68~0.83)、0.84(95%CI:0.78~0.89)、4.9(95%CI:3.5~7.0)、0.28(95%CI:0.21~0.37)、18(95%CI:11~28),绘制SROC曲线,AUC值为0.87,Meta回归显示,测试集数量、模型验证策略、学习模型种类对诊断结果产生的差异无统计学意义。Deek's漏斗图评估提示无发表偏倚,P=0.264。结论基于CT的ML模型鉴别诊断难辨别肾良性肿瘤与肾细胞癌时的敏感性、特异性及AUC值均较高,具有临床推广应用的潜力。Objective To assess the clinical value of CT-based machine learning model in differentiating undistinguishably benign and malignant renal tumor.Methods Studies on the value of CT-based machine learning model for differentiating benign(renal oncocytoma,angiomyolipoma without visible fat)and malignant renal tumor were searched in PubMed,The Cochrane Library,Web of science,Medline,CNKI,and Wanfang from inception to Mar.2022.Data collected were analyzed with Stata 14.0,RevMan 5.4 and meta-Disc 1.4 software.The pooled sensitivity and specificity,positive and negative likelihood ratio,and the diagnostic odds ratio were computed.The summary receiver operating characteristic(SROC)curve was drawn and the area under the curve(AUC)was calculated.The number of test sets,validation strategies and types of learning models were analyzed by subgroup.Meta-regression analysis was performed to explore the source of heterogeneity beyond the threshold effect.Results A total of 12 literatures were included.The pooled sensitivity,specificity,positive likelihood ratio,negative likelihood ratio,and diagnostic odds ratio were 0.76(95%CI:0.68-0.83),0.84(95%CI:0.78-0.89),4.9(95%CI:3.5-7.0),0.28(95%CI:0.21-0.37),18(95%CI:11-28),respectively.The AUC of the SROC curve was 0.87.Subgroup analysis showed that the number of test sets,validation strategies and types of learning models had little effect on the diagnostic results.Deek's funnel plot found no significant publication bias(P=0.264)Conclusions CT-based machine learning model has great sensitivity,specificity and AUC for differentiating undistinguishable benign and malignant renal tumor.It has potential for clinical promotion.

关 键 词：肾肿瘤 肾细胞癌 人工智能 机器学习 计算机断层扫描 肾嗜酸细胞瘤 肾乏脂肪血管平滑肌脂肪瘤 META分析 

分 类 号：R737.11[医药卫生—肿瘤]