探讨CT纹理分析在鉴别体内结石亚型中的价值  

Investigation of the value of using CT texture analysis to differentiate urinary calculi in vivo

作　　者：蔡磊[1] 叶冬晖 潘雪琳[1] 陈剑锋[1] 鲁旭 孙怀强[1] 陈新月 孙家瑜[1] CAI Lei;YE Dong-hui;PAN Xue-lin;CHEN Jian-feng;LU Xu;SUN Huai-qiang;CHEN Xin-yue;SUN Jia-yu(Department of Radiology,West China Hospital,Sichuan University,Chengdu,Sichuan 610041,China;不详)

机构地区：[1]四川大学华西医院放射科,四川成都610041 [2]四川大学华西医院泌尿外科 [3]西门子科研合作,西门子中国

出　　处：《现代预防医学》2021年第13期2488-2492,2496,共6页Modern Preventive Medicine

摘　　要：目的探讨应用CT纹理分析在区分人体内一水草酸钙结石与混合草酸钙结石的价值。方法65例行CT扫描的草酸钙结石患者,以红外光谱分析为标准分为A、B两组:A组单一成分一水草酸钙结石(COM,20例),B组混合草酸钙结石(COM+COD+CaP,45例)。使用ITK-Snap进行图像分割及AK软件提取结石的纹理特征。采用Pearson相关性分析进行特征降维;使用Lasso回归进行建模,以交叉验证方法对回归模型进行检验;绘制受试者工作特征曲线对模型进行诊断表现及实用性检测。结果共提取影像组学特征396个,共线性效应的去除基于Pearson相关性r值,使数据集满足任一两两特征间的r小于0.8,最终纳入特征36个;利用SMOTE方法扩充数据,平衡研究类别不均影像模型构建的问题;数据基于Lasso回归建模,模型调准参数的确定基于5次反复10折内部交叉验证反复采样AUC的平均值,最终在回归模型里系数不为零的纹理特征包括:Entropy、Energy、Nonuniformity、SurfaceArea、Correlation;模型在原始数据的表观诊断表现:AUC为0.9022(95%CI:0.804~1),敏感度、特异度及准确率分别为91.1%,85%和89.2%。模型的普遍适用性检测采用内部反复随机5次抽样法,模型的AUC分别为:0.9671(95%CI:0.9346~0.9996)、0.9489(95%CI:0.904~0.9938)、0.9525(95%CI:0.911~0.9939)、0.9655(95%CI:0.9314~0.9996)、0.9544(95%CI:0.91~0.9989),模型表现稳定。结论基于结石CT图像的纹理特征可以区分体内一水草酸钙结石与混合组分的草酸钙结石。Objective To investigate the value of using CT texture analysis to differentiate urinary calcium calculi in vivo.Methods 65 calcium calculi were analyzed retrospectively and divided into two groups according to the result of infrared spectrometer:group A(COM,20),group B(Mixed,COM+COD+CaP,45),Three-dimensional ROI segmentation was conducted by ITK-SNAP,and AK software was used to extract 396 radiomics features from each segmented ROI in every CT image phase.Pearson correlation analysis was used for feature dimension reduction.The regression model was modeled by Lasso regression,and the regression model was tested by cross validation.ROC curve analysis was performed and the areas under curve(AUC)were calculated for performance evaluation.Results A total of 396 features of image formation were extracted,and 36 of them were texture features reduced by Pearson correlation analysis(r<0.8).SMOTE was used to expand the data and correct the uneven distribution of research category.After Lasso modeling,five features of texture features were finally screened based on internal cross validation,the AUC for the apparent diagnostic performance of the model in the original data was 0.9022(95%CI:0.804-1),and the sensitivity,specificity and accuracy of Lasso regression model were 91.1%,85%and 89.2%,respectively.The general applicability of the model showed a stable performance with AUC:0.9671(95%CI:0.9346-0.9996),0.9489(95%CI:0.904-0.9938),0.9525(95%CI:0.911-0.9939),0.9655(95%CI:0.9314-0.9996),0.9544(95%CI:0.91-0.9989),which was tested by internal repeated random sampling method for 5 times.Conclusion The CT texture parameters is helpful to differentiate COM calculi from other mixed calcium calculi in vivo.

关 键 词：泌尿系结石 纹理分析 计算机断层扫描 

分 类 号：R691.4[医药卫生—泌尿科学]