基于超声影像学特征定量预测甲状腺良恶性的模型构建及应用效果分析

Construction of a model of quantitative predicting benign and malignant thyroid based on the ultrasonic imaging characteristic and the analysis on application effect of that

目的:构建基于超声影像学特征定量预测甲状腺良恶性的模型,并分析其应用效果。方法:选择医院收治的105例甲状腺结节患者,根据病理结果将其分为恶性组(65例)和良性组(40例)。采用蒙特卡罗模拟方法将105例结节样本分为训练集(84例)和预测集(21例)。所有患者均行甲状腺结节常规超声、超声弹性成像(UE)及超声造影检查,记录甲状腺结节钙化、实性成分、回声类型、质地、形态、纵横比、边缘、血流分级、远场、晕、血流位置、被膜等影像特征。利用偏最小二乘-判别分析法(PLS-DA)及Logistic回归分析方法建立甲状腺结节良恶性风险评估预测模型。采用受试者工作特征(ROC)曲线分析两种模型鉴别甲状腺结节良恶性的价值。结果:恶性组患者结节微钙化、实性成分、低回声、形态不均匀、纵横比≥1、边缘不清晰、远场衰减以及被膜受侵比例明显高于良性组(χ^(2)=49.114,χ^(2)=17.960,χ^(2)=24.432,χ^(2)=41.316,χ^(2)=37.813,χ^(2)=39.023,χ^(2)=45.859,χ^(2)=10.033;P<0.05)。ROC曲线分析显示,PLS-DA基于训练集构建模型诊断预测集甲状腺恶性结节的ROC曲线下面积(AUC)为0.948;Logistic回归分析基于训练集构建模型诊断预测集甲状腺恶性结节的AUC为0.901。通过蒙特卡罗模拟运行1 000次分析发现,PLS-DA模型训练集真阳性和真阴性准确率分别为97.11%和99.74%,PLS-DA模型预测集真阳性和真阴性准确率分别为91.68%和96.61%,此时的AUC均值为0.938,95%CI为0.928~0.966。蒙特卡罗模拟运行1000次Logistic回归分析显示,结节微钙化、实性成分、低回声、形态不规则、纵横比≥1、边缘不清晰、远场衰减以及被膜受侵对预测甲状腺肿瘤是否恶化具备统计学意义。结论:基于PLS-DA及Logistic回归分析方法建立的甲状腺结节良恶性风险评估预测模型在鉴别甲状腺结节良恶性的价值较好,其中基于PLS-DA甲状腺结节良恶性预测模型预测价值更高。

Objective: To construct a model that could quantitatively predict the benign and malignant thyroid based on ultrasonic imaging characteristics, and to analyze the application effect of that. Methods: 105 patients with thyroid nodules admitted to the hospital were selected and they were divided into malignant group(65 cases)and benign group(40 cases) according to the pathological results. Monte Carlo simulation method was adopted to divided the 105 samples of nodules into training set(84 cases) and prediction set(21 cases). All patients underwent routine ultrasound, ultrasound elastography(UE) and contrast-enhanced ultrasound examination for thyroid nodules,and the calcification, composition, echo type, texture, shape, aspect ratio, edge, classification of blood flow, far-field,halo, position of blood flow, envelope and other imaging features of thyroid nodules were recorded. The assessment and prediction model of risk for benign and malignant thyroid nodules was established by using partial least-squares discriminant analysis(PLS-DA) and logistic regression analysis, respectively. The receiver operating characteristics(ROC) curve was used to analyze the value of the two models in differentiating benign and malignant thyroid nodules.Results: The proportions of microcalcification, solidity, hypoecho, uneven shape, aspect ratio ≥ 1, unclear edge, far field attenuation and capsule invasion in the malignant group was significantly higher than those in the benign group(x~2= 49.114, x~2=17.960, x~2=24.432, x~2=41.316, x~2=37.813, x~2=39.023, x~2=45.859, x~2=10.033, P<0.05). The results of ROC curve analysis showed that the area under curve(AUC) of the ROC curve of the diagnosing and predicting set of established model on the basis of training set by using PLS-DA was 0.948 for malignant nodules of thyroid, and that of the diagnosing and predicting set of established model on the basis of training set by using logistic regression analysis was 0.901 for malignant nodules of thyroid. The analysis of 1000 times of Monte Carlo simulation run found that the accuracies of true positivity and true negativity of training set of PLS-DA model were 97.11% and 99.74%, and those of prediction set of PLS-DA were 91.68% and 96.61%, respectively. At this time, the AUC mean value was 0.938, and 95% CI=0.928-0.966. The logistic regression analysis of Monte Carlo simulation run with 1000 times showed that the microcalcification of nodule, solidity, hypoecho, uneven shape, aspect ratio ≥ 1, unclear edge, far field attenuation and capsule invasion had statistical significance in predicting whether occurred deterioration of thyroid tumor. Conclusion:The established assessment and prediction models of risk based on PLS-DA and logistic regression analysis for benign and malignant thyroid nodules have better value in identifying benign and malignant thyroid nodules, and the prediction value of the prediction model based on PLS-DA is higher for predicting benign and malignant thyroid nodules.