基于CT影像组学预测感染性肾结石的价值  

Value of the prediction of infectious kidney stones based on CT radiomics

作　　者：蒋旭 曹海明 吴宇[1] 马学平 杨学贞 JIANG Xu;CAO Haiming;WU Yu;MA Xueping;YANG Xuezhen(Department of Urology,The Second Affiliated Hospital of Bengbu Medical College,Bengbu,Anhui 233040,China)

机构地区：[1]蚌埠医学院第二附属医院泌尿外科,安徽蚌埠233002

出　　处：《安徽医药》2021年第7期1401-1406,共6页Anhui Medical and Pharmaceutical Journal

基　　金：蚌埠医学院自然科学类项目(BYKY2019164ZD)。

摘　　要：目的利用计算机辅助诊断技术探讨影像组学数据在鉴别结石成分的应用。方法回顾性分析2018年12月至2020年12月蚌埠医学院第二附属医院140例采用经皮肾镜取石术(PCNL)治疗的病人,收集临床治疗前因素及CT图像的相关数据,术后通过傅里叶转换红外光谱法测定结石成分,并以结石主成分≥70%定义结石性质。利用图像分析软件和计算机程序设计语言工具,从每位病人的CT图像感兴趣区域中提取出105个影像组学特征。按照7∶3将数据分为训练集98例和验证集42例,采用t检验、χ^(2)检验、秩和检验及LASSO回归分析法对训练集进行变量选择,得到最佳特征选集,最终利用R语言软件构建感染性结石的列线图预测模型,模型评价指标为分辨度和符合度,采用ROC曲线下面积评价模型的分辨度,绘制校正曲线评价模型的符合度,外部验证利用验证集数据评估模型效果并绘制ROC曲线。结果140例病人术后结石成分分析报告显示感染性结石49例,非感染性结石91例。训练集数据临床治疗前因素的分析结果显示差异有统计学意义(P<0.05)为女性、尿蛋白、尿碱性、尿亚硝酸盐、尿培养、尿白细胞。感染性结石组尿白细胞数为162(46.5,944.0)个,非感染性结石组56(10.5,169.5)个,P=0.003;感染性结石组尿酸为(285.22±83.22)μmol/L,非感染性结石组(324.85±99.95)μmol/L,P=0.046。训练集影像组学LASSO回归分析法筛选得到8个相关性较高的影像组学特征变量。进一步logistic多因素分析得出最佳特征选集包括2个临床治疗前因素(女性、尿碱性)和2个影像组学特征(群集阴影、大依赖性低灰度级强度),训练集AUC(0.892,95%CI:0.830~0.954),验证集AUC(0.842,95%CI:0.702~0.981),校正曲线表明模型符合度较好。结论计算机辅助诊断技术帮助下提取的影像组学特征,结合临床治疗前因素,有助于术前判断感染性肾结石的发生风险。Objective To explore the application of imaging data in differentiating stone components by computer-aided diagnosed.Methods Bengbu Medical College from December 2018 to December 2020 were retrospectively analyzed,the data on pre-clinical factors and computed tomography(CT)images were collected,the calculi specimens were investigated by Fourier transform infrared spectroscope(FTIR)after the operation,and the stone character was defined by stone major components≥70%.Using image analysis software and computer programming language tools,extraction of 105 imaging features from regions of CT images of interest for each patient.The data were assigned into 98 training sets and 42 testing sets according to 7:3.The several statistical analysis methods(t test,chi-square test,rank sum test and lasso regression analysis)were used to select the variables of the training set,and the best feature selection was obtained.R language software was adopted to construct a prediction model for the histogram of infectious stones,the evaluation indexes of the model were resolution and conformity,the area under Receiver Operating Characteristic(ROC)Curve was used to evaluate the resolution of model,the correction curve was drawn to evaluate the conformity of model,external validation evaluated the model effect using validation set data and ROC curve was drew.Results tious stones and 91 cases of non-infectious stones.Analysis of pre-clinical factors in training set data showed that there were significant differences in female,urine protein,alkaline urine,urine nitrite,urine culture,white blood cells in urine and number of white blood cells in urine(P<0.05).The number of white blood cells in urine infectious stone group and non-infectious stone group was 162(46.5,944.0)and 56(10.5,169.5),respectively(P=0.003).The uric acid in urine infectious stone group and non-infectious stone group was(285.22±83.22)μmol/L and(324.85±99.95)μmol/L,respectively(P=0.046).Eight high correlation radiomics feature variables were obtained by training set a

关 键 词：尿路结石症 影像组学 泌尿道感染 电子计算机断层扫描 感染性结石 列线图 

分 类 号：R692.4[医药卫生—泌尿科学]