甲状腺实性结节超声诊断的Logistic回归分析

摘要

Objective To evaluate the application of binary Logistic regression model in analyzing the ultrasonographic indexes of solid thyroid nodules. Methods The indexes of two-dimensional gray scale ultrosonography, two-dimensional color Doppler flow imaging and ultrasonic elastography were evaluated in 118 thyroid lesions, which were confirmed by pathological diagnosis after surgery. A Logistic regression model for predicting thyroid malignancy on the basis of ultrosonographic indexes were developed. Results Three ultrasonic indexes were finally entered into the Logisic regression model including elasticity score, elasticity strain ratio and shape. The percentage correction of prediction was 86. 4% ( 102/118 ). The area under the ROC curve was 0. 931. Conclusions The multivariate analysis model of binary Logistic regression can describe and analyze the process of differential diagnosis of malignant and benign solid thyroid nodules by ultrosonography, and select out the valuable indexes for differential diagnosis. Ultrasound elastography plays an important role in differentiating benign and malignant solid thyroid nodules.%目的 应用二分类Logistic回归模型分析甲状腺实性结节超声检查各指标对良恶性结节鉴别诊断的意义.方法 总结经手术病理证实的118个甲状腺结节术前的二维灰阶、彩色多普勒及弹性成像的各项超声诊断指标,采用多因素回归分析方法,建立Logistic回归模型.结果 共有3个变量进入回归方程,分别为弹性评分、弹性应变率和结节形态,该Logistic模型对甲状腺结节良恶性预报的正确率为86.4%(102/118),ROC曲线下面积为0.931.结论 二分类Logistic回归模型分析能筛选出对甲状腺实性结节有鉴别诊断意义的特征变量及弹性成像方法所反应出来的硬度信息,在良恶性结节的鉴别诊断中具有较高的临床应用价值.