基于腹腔镜超声的纹理分析鉴别最大径≤4 cm的肾透明细胞癌与肾血管平滑肌脂肪瘤的价值

Value of texture analysis based on laparoscopic ultrasound in differentiating renal clear cell carcinoma and renal angiomyolipoma with maximum diameter≤4 cm

目的评价基于腹腔镜超声的纹理分析在鉴别最大径≤4 cm的肾透明细胞癌与肾血管平滑肌脂肪瘤的诊断价值。方法回顾性分析我院经病理证实的124例最大径≤4 cm的肾脏占位病灶(肾透明细胞癌患者66例和肾血管平滑肌脂肪瘤患者58例)的腹腔镜超声检查资料,将其随机分为训练组87例和验证组37例。通过ITK-SNAPE软件手工勾画感兴趣区,采用Pyradiomics工具包提取纹理特征。使用组内相关系数(ICC)选择具有良好稳定性和可重复性的特征;使用最大相关最小冗余(mRMR)和最小绝对收缩和选择算子(LASSO)回归方法进行特征选择和建模,采用受试者工作特征(ROC)曲线评估该模型鉴别肾透明细胞癌与肾血管平滑肌脂肪瘤的诊断效能。结果预测模型由2个纹理特征构成,分别为glszmGrayLevelNonUniformity和firstorder90 Percentile,构建模型的计算公式为:Y=0.185+0.387×glszmGrayLevelNonUniformity-0.319×firstorder90 Percentile。在训练组和验证组中鉴别肾透明细胞癌与肾血管平滑肌脂肪瘤的ROC曲线下面积、敏感性、特异性分别为0.911、0.891、0.854和0.900、0.950、0.647。结论基于腹腔镜超声的纹理分析可以准确鉴别最大径≤4 cm的肾透明细胞癌与肾血管平滑肌脂肪瘤。

ObjectiveTo evaluate the diagnostic value of texture analysis based on laparoscopic ultrasonography in differentiating renal clear cell carcinoma from renal angiomyolipoma with maximum diameter≤4 cm.MethodsThe data of 124cases of small renal space occupying lesions(maximum diameter≤4 cm,including 66 cases of renal clear cell carcinoma and 58cases of renal angiomyolipoma)confirmed by pathology were retrospectively collected.They were randomly divided into training group(n=87)and verification group(n=37).Area of interesting was drawn by ITK-SNAP software and the texture features were extracted by python-based Pyradiomics package.Intra-correlation coefficients were used to select the features with excellent stability and repeatability.The minimum-redundancy maximum-relevance(m RMR) and the least absolute shrinkage and selection operator(LASSO)algorithms were used for feature selection and model development for discriminating renal clear cell carcinoma from renal angiomyolipoma.Receiver operating characteristic(ROC) curve analysis was used to evaluate the diagnostic efficacy of the model.ResultsThe best texture feature combination were the glszmGray Level Non Uniformity and the firstorder90 Percentile.The calculation formula for the model was Y=0.185+0.387×glszmGray Level Non Uniformity-0.319×firstorder90 Percentile.The area under ROC curve,sensitivity and specificity were 0.911,0.891,0.854 in the training group and0.900,0.950,0.647 in the verification group,respectively.ConclusionTexture analysis based on laparoscopic ultrasound can accurately identify renal clear cell carcinoma and renal angiomyolipoma with maximum diameter≤4 cm.