多模式超声检查诊断甲状腺癌患者的Logistic回归分析

Logistic regression analysis of multi-mode ultrasonography in the diagnosis of thyroid carcinoma

目的探讨灰阶超声(GSU)、超声造影(CEUS)及实时弹性成像(EI)多模式成像在甲状腺实性结节良恶性鉴别诊断中的价值。方法选取2011年9月至2014年11月首都医科大学宣武医院收治的甲状腺癌患者120例,共124个实性结节,其中良性结节49个,恶性结节75个。对多模式成像资料行回顾分析,并与手术病理诊断结果比较,甲状腺良性结节与恶性结节组间超声声像图特征的比较采用χ2检验,将有统计学意义的参数作为自变量X,以病理结果良、恶性作为因变量Y,按常规超声检查可能出现的组合形式设立4个诊断模型,模型Ⅰ为灰阶超声特征;模型Ⅱ为灰阶超声+超声造影特征;模型Ⅲ为灰阶超声+弹性成像特征;模型Ⅳ为灰阶超声+超声造影+弹性成像特征;建立Logistic回归模型分析,建立ROC曲线,计算曲线下面积(AUC)并进行多因素Logistic回归分析。结果甲状腺良、恶性结节超声声像图特征中纵横比、边缘情况、回声类型、超声造影特征、弹性成像评分及弹性应变率比值等参数比较,差异有统计学意义(χ2=13.966、7.106、24.215、17.250、25.494、27.158,P均<0.05)。4个Logistic回归模型经检测均有统计学意义(χ2=5.37、5.37、8.15、8.15,P均<0.05),模型Ⅰ﹑Ⅱ﹑Ⅲ﹑Ⅳ恶性病变检出的敏感度为73.8%、75.9%、79.7%、83.3%,特异度为67.5%、70.7%、85.2%、86.7%,准确率71.8%、74.2%、81.8%、84.4%。4种模型的Logistic方程预测值ROC曲线下面积为0.778、0.809、0.875、0.898(P均<0.01)。多因素哑变量分析,甲状腺结节的纵横比、边缘及弹性应变率比值是诊断甲状腺结节的有效指标(χ2=13.73、7.541、11.461,P均<0.05)。结论多模式超声检查技术联合应用在甲状腺实性肿瘤的良恶性鉴别中准确性具有较高的诊断价值。

Objective To evaluate the diagnostic value of gray-scale ultrasonography （GSU） combined with contrast-enhanced ultrasonography （CEUS） and real-time elastography imaging （EI） in solid thyroid nodules. Methods A total of 124 solid nodules were collected from Xuanwu Hospital of Capital Medical University from September 2011 to November 2014, including 49 benign nodules and 75 malignant nodules confirmed by surgical pathology. The data of multimodal imaging were retrospectively analyzed and compared with the pathological results. The parameters among the groups were compared by A3 test. Four diagnostic models including the statistically significant independent variables were established accord to different combinations of routine ultrasound examination. The binary Logistic regression equation using the pathological results as the dependent variable Y and the ultrasound features of thyroid nodules as independent variable X was established for calculating the area under the receiver operator curve （ROC） and multiple regression analysis furthermore. Results There were statistically significant differences in aspect ratio, margin condition, echo type, CEUS features, elastic imaging score and elastic strain rate of benign and malignant thyroid nodules （χ^22=13.966, 7.106, 24.215, 17.250 and 27.158, P 〈 0.05） .The four Logistic regression models were statistically significant （χ^2=5.37, 5.37, 8.155, 8.155, all P 〈 0.05）. By using model Ⅰ, Ⅱ, Ⅲ and Ⅳ, the diagnostic sensitivity, specificity and accuracy of malignant nodules were （73.8%, 75.9%, 79.7%, 83.3%）, （67.5%, 70.7%, 85.2%, 86.7%）, and （71.8%, 74.2% 81.8%, 84.4%） respectively. The area under the ROC of logistic equation predicted values were （0.778, 0.809, 0.875 and 0.898, all P 〈 0.01）. Multivariate categorical analysis showed that aspect ratio, margin condition and the elastic strain rate ratio （χ^2=13.73, 7.541 and 11.461, all P 〈 0.05） of thyroid nodules were effective diagnostic indicators. Conclusions Our study indicated that CEUS and EI could be helpful modality for GSU to improve the diagnostic accuracy of thyroid nodules. The combined application of multi mode ultrasound examination has a high diagnostic value.