摘要
目的基于CT增强评估不同范围瘤周组织的影像组学特征分析,验证其术前鉴别肾乏脂肪血管平滑肌脂肪瘤(Fat-Poor Angiomyolipoma,fp_AML)与肾透明细胞癌(Clear Cell Renal Cell Carcinoma,ccRCC)的效能。方法回顾性分析150例直径≤4 cm的肾脏肿瘤患者术前CT增强扫描图像,包括ccRCC103例,fp_AML 47例。在皮髓质期CT增强图像上手工勾画出肿瘤区域、不同范围瘤周区域以及肿瘤+瘤周模型的三维感兴趣区(Volume of Interests,VOIs),按照7∶3的比例划分训练集与测试集,提取并筛选影像组学特征后,建立逻辑回归模型(Logistics Regression),通过受试者工作特征曲线评价各个模型对于区分≤4 cm ccRCC与fp_AML鉴别诊断的能力。结果在6个模型中,表现最佳的是肿瘤体积(Tumor Mass Volume,TMV)+肿瘤边缘外2 mm(Peritumoral Volume,PTV0~2)联合模型,训练集的AUC为0.990、特异度为0.96、灵敏度为0.88、准确度为0.93;验证集的AUC为0.931、特异度为0.87、灵敏度为0.71、准确度为0.82。在4个非联合模型中,肿瘤边界内外2 mm(PTV2~2)模型的鉴别效能明显优于其他3个单独模型,训练集的AUC为0.911、特异度为0.92、灵敏度为0.70、准确度为0.85;验证集的AUC为0.917、特异度为0.97、灵敏度为0.71、准确度为0.89。结论在基于CT的单个模型中,PTV2~2瘤周模型表现最好,即PTV2~2的瘤周范围更能够有效鉴别fp_AML和ccRCC;联合肿瘤及瘤周模型中,TMV+PTV0~2联合模型性能最佳,可更准确、全面地反映肿瘤的特征和异质性,因此可作为鉴别fp_AML和ccRCC最有效能的方法。
Objective To verify the effectiveness in preoperative differentiation between fat-poor angiomyolipoma(fp_AML)and clear cell renal cell carcinoma(ccRCC)by analyzing radiomic features based on CT-enhanced assessment in different ranges of peritumoral tissue.Methods The preoperative CT-enhanced scanning images of 150 patients with renal tumors with diameter≤4 cm were analyzed retrospectively,including 103 cases of ccRCC and 47 cases of fp_AML.The tumor area,different ranges of peritumor areas and the volume of interests of the tumor and peritumor model were manually delineated on the CT-enhanced images in corticomedullary phase.The training and test sets were divided according to the ratio of 7∶3.After extracting and screening the radiomics features,a Logistic regression model was established.The effectiveness of each model to differentiate ccRCC from fp_AML was evaluated by using the receiver operating characteristic curve.Results Among the six models,the best performing model was the combined tumor mass volume(TMV)and 2 mm beyond the edge of the tumor(peritumoral volume,PTV0~2)model with the area under curve(AUC)of 0.990,specificity of 0.96,sensitivity of 0.88 and accuracy of 0.93 for the training data,and AUC of 0.931,specificity of 0.87,sensitivity of 0.71 and accuracy of 0.82 for the validation model.Among the four non-combined models,the discrimination efficiency of PTV2~2(2 mm inside and outside the tumor boundary)model significantly outperformed the remaining three individual models,with AUC of 0.911,specificity of 0.92,sensitivity of 0.70,and accuracy of 0.85 for the training data and an AUC of 0.917,specificity of 0.97,sensitivity of 0.71,and accuracy of 0.89 for the validation set.Conclusion Among the CT-based individual models,the PTV2~2 peritumor model performs is best,that is,the peritumoral range of PTV2~2 is more effective in identifying fp_AML and ccRCC;among the combined tumor and peritumor models,the combined TMV and PTV0~2 model performs best,which can more accurately and comprehensively refl
作者
贺琬淋
方维东
HE Wanlin;FANG Weidong(Department of Radiology,The First Affiliated Hospital of Chongqing Medical University,Chongqing 400016,China)
出处
《中国医疗设备》
2022年第8期123-127,共5页
China Medical Devices
