应用Logistic回归分析方法研究甲状腺良恶性结节的超声特征

发布时间：2018-10-23 20:44

【摘要】：目的通过建立Logistic回归模型研究对甲状腺良恶性结节鉴别诊断有意义的超声特征。方法 173个甲状腺结节经手术病理证实,术前均记录灰阶和血流超声特征并对其进行评分;利用Logistic逐步回归分析筛选出甲状腺结节良恶性鉴别诊断的独立特征指标;绘制ROC曲线,评价Logistic回归模型的预报能力。结果 1运用二分类Logistic回归分析,形状、内部回声、纵横比、周边晕、钙化进入回归模型,并建立回归方程式,该回归模型对甲状腺良恶性结节预报的准确率为90.7%。2恶性可能性=1/1+e-Z,Z=bo+b1x1+……+bpxp(e值为2.718)。3ROC曲线下面积为0.822(P0.001),说明该模型的拟合效果较好。结论二分类Logistic回归分析能够筛选出对甲状腺结节良恶性鉴别诊断有意义的特征变量,有助于判断甲状腺结节的良恶性。
[Abstract]:Objective to establish a Logistic regression model for the differential diagnosis of benign and malignant thyroid nodules. Methods 173 thyroid nodules confirmed by operation and pathology were recorded and graded with gray scale and blood flow ultrasound before operation, and the independent characteristic indexes of differential diagnosis of benign and malignant thyroid nodules were screened by Logistic stepwise regression analysis. The ROC curve is drawn to evaluate the prediction ability of Logistic regression model. Results 1 the regression model was established by using two-classification Logistic regression analysis, shape, internal echo, aspect ratio, peripheral halo and calcification, and the regression equation was established. The accuracy of the regression model for the prediction of benign and malignant thyroid nodules is 90.7.2 the probability of malignancy = 1 / 1 e-ZZN bo b1x1. Bpxp (e value is 2.718). The area under the 3ROC curve is 0.822 (P0.001), which shows that the fitting effect of the model is good. Conclusion the two classification Logistic regression analysis can screen out the characteristic variables which have significance in the differential diagnosis of benign and malignant thyroid nodules, which is helpful to judge the benign and malignant thyroid nodules.
【作者单位】：山西医科大学第一医院超声科;
【基金】：山西省回国留学人员科研资助项目(2014-073) 山西省高校131领军人才工程优秀拔尖人才支持基金资助项目(2013-404)
【分类号】：R445.1;R581

【相似文献】