应用Logistic回归模型评价超声及超声造影特征对甲状腺部分囊性结节（PCTN）良恶性的鉴别诊断价值。

选取2016年1月至2018年12月在解放军总医院第一医学中心就诊的135例PCTN患者的临床资料和超声图像进行回顾性分析，共159个结节，良性结节109个，其中结节性甲状腺肿94个，腺瘤15个；恶性结节50个，其中乳头状癌47个，髓样癌2个，未分化癌1个。以病理结果作为因变量，各超声特征作为自变量，建立二分类Logistic回归模型，并评价其预测结节良恶性能力。

共4个特征进入Logistic回归模型，分别是钙化、边界、实性部分的形态和增强均匀度。Logistic回归模型对甲状腺部分囊性结节良恶性预测的敏感度、特异度、准确性分别为86.0%、94.5%、80.5 %，受试者工作特征曲线下面积为0.902（95%可信区间为0.845~0.944）。

常规超声联合超声造影特征的Logistic回归模型有助于PCTN良恶性鉴别诊断。

To assess the value of conventional ultrasound combined with contrast-enhanced ultrasound in the differential diagnosis of benign and malignant partially cystic thyroid nodules (PCTNs).

The clinical data and ultrasound images of 135 patients with 159 PCTNs who were treated at the First Medical Center of PLA General Hospital from January 2016 to December 2018 were analyzed retrospectively. There were 109 benign nodules, including 94 nodular goiters and 15 adenomas, and 50 malignant nodules, including 47 papillary carcinomas, 2 medullary carcinoma, and 1 undifferentiated carcinoma. A binary logistic regression model was established with pathological results as the response variables and ultrasonic characteristics as the independent variables, and the ability of the model for predicting benign and malignant nodules was evaluated.

A total of four features were entered into the logistic regression model, including calcification, boundary, solid part morphology, and enhanced uniformity. The sensitivity, specificity, and accuracy of the logistic regression model were 86.0%, 94.5%, and 80.5% respectively, and the area under the ROC curve was 0.902 (95% confidence interval: 0.845-0.944).

The logistic regression model of routine ultrasound combined with contrast-enhanced ultrasound is helpful in the differential diagnosis of benign and malignant thyroid PCTNs.