logistic回归模型在甲状腺良恶性结节超声诊断中的价值

The value of logistic regression model in ultrasound diagnosis of benign and malignant thyroid nodules

目的应用logistic回归模型筛选出能够鉴别甲状腺结节良恶性的声像图特征并建立回归模型,评估回归模型在甲状腺良恶性鉴别中的诊断价值。方法回顾性分析175例经手术病理证实的甲状腺良恶性结节的临床资料和常规超声特征,应用单因素及多因素二元logistic回归分析评价临床一般资料、常规超声特征在鉴别甲状腺良恶性结节中的价值。结果2组患者的性别、年龄、肿块内部回声均匀性、血液供应程度、血液供应模式及有无合并弥漫性甲状腺病变比较,差异均无统计学意义(P值均>0.05);结节的直径、内部回声、回声水平、边界、边缘、内部微钙化情况、纵横比、后方有无声衰减、甲状腺外包膜有无侵犯组间比较,差异均有统计学意义(P值均<0.05);logistic多因素回归分析显示:微钙化对甲状腺结节的鉴别有统计学意义(OR=3.280,95%CI:1.031~10.435,P<0.05);结节纵横比对甲状腺结节的鉴别有统计学意义(OR=3.653,95%CI:1.013~13.174,P<0.05);结节边缘对甲状腺结节的鉴别有统计学意义(OR=5.731,95%CI:1.995~16.460,P<0.01)。以手术病理结果及回归模型预测甲状腺恶性病变的概率绘制ROC曲线,结果显示ROC曲线下面积为0.895(95%CI:0.845~0.944)。结论甲状腺良恶性结节具有一定的超声特征性,logistic回归模型有较高的诊断价值。

Objective To use the logistic regression model to screen out the sonographic features that can distinguish the benign and malignant thyroid nodules and establish a regression model to evaluate the diagnostic value of the regression model in the differentiation of benign and malignant thyroid nodules.Methods The clinical data and conventional ultrasound features of 175 cases of benign and malignant thyroid nodules confirmed by surgery and pathology were retrospectively analyzed.The value of clinical general data and conventional ultrasound features in differentiating benign and malignant thyroid nodules was evaluated by single factor and multiple factor binary logistic regression analysis.Results There was no significant difference between the two groups in gender,age,internal echo uniformity,blood supply degree,blood supply pattern,and presence or absence of diffuse thyroid disease(P>0.05);the diameter,internal echo,echo level,boundary,edge,internal microcalcification,aspect ratio,silent attenuation at the back,and invasion of the outer capsule of the thyroid gland were significantly different between the two groups(P<0.05);logistic multivariate regression analysis showed that microcalcification had statistical significance in the differentiation of thyroid nodules(OR=3.280,95%CI:1.031-10.435,P<0.05);the aspect ratio of thyroid nodules was statistically significant(OR=3.653,95%CI:1.013-13.174,P<0.05);the edge of the nodule had statistical significance in the differentiation of thyroid nodules(OR=5.731,95%CI:1.995-16.460,P<0.01).The ROC curve was drawn according to the surgical pathological results and the probability of predicting thyroid malignancy by regression model.The results showed that the area under the ROC curve was 0.895(95%CI:0.845-0.944).Conclusions The benign and malignant thyroid nodules have certain ultrasonic characteristics,and the logistic regression model has high diagnostic value.