影像组学在鉴别伴结石肾积水是否伴发肾细胞癌中的应用

A radiomic approach to differential diagnosis of renal cell carcinoma in patients with hydronephrosis and renal calculi

目的利用有监督的机器学习方法探讨影像组学分析在鉴别伴结石肾积水是否伴发肾细胞癌中的应用。方法回顾性分析经病理确诊的66例伴结石肾积水患者的腹部CT扫描,其中31例伴发肾细胞癌。对每位患者的三维肿瘤区域提取18个非纹理特征和344个纹理特征,并应用无限特征选择技术(InfFS)结合支持向量机分类器的方法(SVM)进行特征选择。最后将最佳特征子集训练SVM分类器并对伴结石肾积水是否伴发肾细胞癌进行预测。结果 12个纹理特征入选最佳特征子集,且SVMInfFS对伴结石肾积水是否伴发肾肿瘤的预测结果如下:感受曲线下面积、准确率、敏感性、特异性、假阳性和假阴性分别为0.907、81.0%、70.0%、90.9%、9.1%和30.0%。临床医生以分类结果作为辅助信息进行诊断的结果如下:准确率、敏感性、特异性、假阳性和假阴性分别为90.5%、80.0%、100%、0.00%、20.0%。结论基于有监督机器学习的计算机辅助分类模型,可有效提取的辅助诊断信息,提高伴结石肾积水是否伴发肾细胞癌的诊断率。

Objective To explore the application of radiomic analysis in differential diagnosis of renal cell carcinoma in patients with hydronephrosis and renal calculi using supervised machine learning methods.Method The abdominal CT scan data were retrospectively analyzed for 66 patients with pathologically confirmed hydronephrosis and renal calculi,among whom 35 patients had renal cell carcinoma.In each case 18 non-texture features and 344 texture features were extracted from the region of interest(ROI).Infinite feature selection(InfFS)-based forward feature selection method coupled with support vector machine(SVM)classifier was used to select the optimal feature subset.SVM was trained and performed the prediction using the selected feature subset to classify whether hydronephrosis with renal calculi was associated with renal cell carcinoma.Results A total of 12 texture features were selected as the optimal features.The area under curve(AUC),accuracy,sensitivity,specificity,false positive rate and false negative rate of the SVM-InfFS model for predicting accompanying renal tumors in patients with hydronephrosis and calculi were 0.907,81.0%,70.0%,90.9%,9.1%,and 30.0%,respectively.The diagnostic accuracy,sensitivity,specificity,false positive and false negative rates by the clinicians provided with these classification results were 90.5%,80.0%,100%,0.00%,and 20.0%,respectively.Conclusion The computer-aided classification model based on supervised machine learning can effectively extract the diagnostic information and improve the diagnostic rate of renal cell carcinoma associated with hydronephrosis and renal calculi.