应用高频和彩色多普勒超声鉴别甲状腺结节良恶性的Logistic回归模型

A Logistic Regression Model for Differential Diagnosis of Benign and Malignant Thyroid Nodules by High Frequency and Color Doppler Ultrasound

目的:建立应用高频和彩色多普勒超声对甲状腺结节良恶性鉴别诊断的Logistic回归模型。材料与方法:回顾性分析经病理证实的119例甲状腺结节患者,其中良性68例,恶性51例。比较两组灰阶及彩色多普勒声像特征,对其超声特征进行多因素回归分析,建立二分类Logistic回归模型,绘制ROC曲线并计算曲线下面积。结果:甲状腺良性结节似圆度较恶性组偏小(P<0.01),恶性结节边界欠清,有微钙化、无声晕或声晕欠完整、后方回声衰减,回声表现为低回声及血流形态者Ⅱ级及Ⅲ级者检出率较良性组高(P<0.01),恶性结节RI高于良性(P<0.05),两组之间PSV差异无统计学意义(P>0.05)。多因素回归分析显示最后进入Logistic模型的4个特征分别为似圆度、微钙化、声晕及血流分布形态,ROC曲线下面积为0.957。结论:应用高频和彩色多普勒超声对甲状腺结节良恶性鉴别的Logistic回归模型有助于鉴别甲状腺良、恶性结节。

Purpose To establish a logistic regression model for differential diagnosis of benign and malignant thyroid nodules by high frequency and color doppler ultrasound.Materials and Methods One hundred and nineteen cases of thyroid nodules diagnosed pathologically were analyzed retrospectively,sixty-eight nodules were benign,and the remaining fifty-one nodules were malignant.The ultrasonographic parameters of benign and malignant masses were recorded and compared using multiple stepwise binary logistic regression analysis.The ROC curve was drawn and the areas under curves were calculated.Results Circularity of benign nodules was significantly higher than malignant nodules（P〈0.01）.Malignant nodules manifested more microcalcification,absence of halo,unclear edge,posterior echo attenuation,hypoechoic mass,Grade Ⅱ and Ⅲ morphologic feature of blood flow（P〈0.01）,and higher RI of blood flow compared with benign nodules（P〈0.05）.There was no statistically significant difference between benign and malignant nodules in PSV（P〉0.05）.Four ultrasonic features were finally implemented into the Logistic regression model including：circularity,microcalcification,halo and the morphologic feature of blood flow,and the area under the ROC curve was 0.957.Conclusion The logistic regression model can be helpful for differentiation of benign or malignant thyroid nodules.