CT影像与纹理特征联合模型在甲状腺良恶性结节鉴别中的价值

The Value of Combination Model based on Texture Analysis Combined with CT Finds in Differentiating Benign and Malignant Thyroid Nodules

目的 比较甲状腺良恶性结节CT影像征象及纹理特征差异,构建两者鉴别诊断模型。方法 回顾性分析110例经手术病理证实分类甲状腺结节患者的125个结节,其中良性组66个,恶性组59个,提取每个结节CT征象与平扫、动脉期及静脉期3个期相的纹理特征。采用χ;检验筛选良恶性结节有差异的影像特征,进一步采用Logistic回归分析构建单纯影像模型。对结节纹理特征筛选、降维,建立剩余纹理参数的受试者工作特征曲线(ROC),获取ROC曲线下面积(AUC)。使用筛选后的常规影像特征及纹理特征组合建立联合Logistic回归模型并获得相应AUC。结果 影像模型的AUC为0.821,灵敏度为66.10%,特异度为81.82%。纹理特征中,静脉期方差具有最高鉴别效能,其AUC为0.696。由以上4个特征构建的联合模型AUC为0.845,灵敏度为76.27%、特异度为83.33%。联合模型的AUC、灵敏度及特异度均高于影像模型。结论 综合CT影像特征及纹理特征的基础上建立联合模型,有助于甲状腺良恶性结节鉴别诊断。

Objective To compare the differences of CT imaging characteristics and texture parameters between benign and malignant thyroid nodules, and furthermore, to develop the differentiation models.Methods A total of 125 nodules confirmed by pathological including 66 benign and 59 malignant in110 patients were retrospectively analyzed. CT imaging characteristics and texture parameters for each nodule in un-enhancement, arterial and venous phase were obtained. Chi-square test were performed to select image features which show different between benign and malignant nodules,and logistic regression analysis was conducted to build image model. After filtered the texture parameters, receiver operating characteristic(ROC) curve analyses and area under the ROC curve(AUC)were performed for the remaining features to determine their diagnostic accuracy in differentiating malignant and benign lesions. Then, the combination Logistic regression model was developed using texture parameter with highest AUC and selected image features. Results The pure image model achieved an AUC of 0.821, with sensitivity of 66.10% and specificity of 81.82%. Among the texture features, the variance in venous phase shown the highest AUC of 0.696. The combined model yielded an AUC, sensitivity and specificity of 0.845, 76.27% and 83.33%, respectively. Compared with image model, the AUC, sensitivity and specificity was improved in combined model. Conclusion The combined model based on conventional CT characteristics and texture features have promising potential for diagnostic differentiation of benign and malignant thyroid nodules.