超声影像组学在腮腺多形性腺瘤与腺淋巴瘤鉴别诊断中的应用价值

Application value of ultrasound imaging radiomics in differential diagnosis of parotid pleomorphic adenoma and adenolymphoma

目的 探讨超声影像组学在腮腺多形性腺瘤与腺淋巴瘤鉴别诊断中的应用价值。方法 选取我院经手术病理证实的133例腮腺多形性腺瘤和99例腺淋巴瘤的超声图像,按7∶3的比例分为训练集162例和验证集70例,其中训练集中腮腺多形性腺瘤101例、腺淋巴瘤61例,验证集中腮腺多形性腺瘤32例、腺淋巴瘤38例。采用ITK-SNAP软件手动勾画训练集肿瘤的感兴趣区,Pyradiomics软件提取二维超声腮腺病灶影像组学定量特征,将提取的特征正则化,采用Spearman相关分析、最大相关最小冗余(mRMR)算法及最小绝对收缩和选择算子(LASSO)回归分析筛选最佳特征。分别采用支持向量机(SVM)、K紧邻(KNN)、决策树(Decision Tree)3种机器学习算法根据筛选的最佳特征构建超声影像组学模型。绘制受试者工作特征(ROC)曲线分析3种模型鉴别训练集及验证集中腮腺多形性腺瘤与腺淋巴瘤的诊断效能。应用决策曲线分析(DCA)评估模型的临床应用价值。结果 训练集中,SVM、KNN、Decision Tree 3种机器学习算法构建的超声影像组学模型鉴别腮腺多形性腺瘤与腺淋巴瘤的AUC分别为0.893、0.825、1.000;验证集中,SVM、KNN、Decision Tree 3种机器学习算法构建的超声影像组学模型鉴别腮腺多形性腺瘤与腺淋巴瘤的AUC分别为0.848、0.721、0.620,其中SVM算法构建的超声影像组模型AUC高于其他两种算法所构建的模型,差异均有统计学意义(均P<0.05)。DCA显示SVM算法构建的超声影像组学模型的临床获益效能最好。结论 超声影像组学在腮腺多形性腺瘤与腺淋巴瘤鉴别诊断中有一定的应用价值;其中以SVM算法构建的超声影像组学模型诊断效能最佳。

Objective To investigate the application value of ultrasound imaging radiomics in the differential diagnosis of parotid pleomorphic adenoma and adenolymphoma.Methods The ultrasonic images of 133 cases of parotid pleomorphic adenoma and 99 cases of adenolymphoma confirmed by surgical pathology were retrospectively collected.According to the ratio of 7∶3,there were 162 cases in training set and 70 cases in verification set,the training set included 101 parotid pleomorphic adenomas and 61 adenolymphoma,and the verification set included 32 parotid pleomorphic adenomas and 38 adenolymphoma.The region of interest of the tumor was manually delineated on ITK-SNAP software,and the quantitative features of two-dimensional ultrasonic parotid focus imaging were extracted by Pyradiomics software,the extracted features were regularized.Spearman correlation analysis,maximum relevance minimum redundancy(mRMR),the least absolute shrinkage and selection operator(LASSO)model were used to select the optimal features.3 machine learning algorithms,namely support vector machine(SVM),K-nearest neighbor(KNN)and Decision Tree were employed to construct ultrasound radiomics models based on the selected features.Receiver operating characteristic(ROC)curves were drawn to evaluate the diagnostic efficiency of the models in differential diagnosis of parotid pleomorphic adenoma and adenolymphoma in the training set and verification set.Decision curve analysis(DCA)was used to evaluate the clinical application value of the models.Results In training set,the area under the curve(AUC)of the ultrasound imaging radiomics model constructed by SVM,KNN and Decision Tree were 0.893,0.825,1.000 in the differential diagnosis of parotid pleomorphic adenoma and adenolymphoma.In verification set,AUC of the ultrasound imaging radiomics model constructed by SVM,KNN and Decision Tree were 0.848,0.721,0.620,60.8%,respectively.The AUC of the ultrasound imaging radiomics model constructed by SVM algorithm was higher than that of the other two algorithms,and the difference were statistically significant(both P<0.05).DCA showed that the SVM algorithm had best clinical benefit and efficiency.Conclusion Ultrasound imaging radiomics has a certain application value in the differential diagnosis of parotid pleomorphic adenoma and adenolymphoma,with the SVM algorithm providing the best diagnostic performance.