基于超声影像组学的预测模型早期诊断甲状腺微小乳头状癌淋巴结转移的临床价值

Clinical value of ultrasound imageomics-based predictive model for early diagnosis of lymph node metastasis in papillary thyroid microcarcinoma

目的基于超声影像组学联合临床特征构建预测模型,探讨其诊断甲状腺微小乳头状癌(PTMC)患者颈部淋巴结转移的临床价值。方法选取我院经术后病理证实的PTMC患者184例,按7∶3比例随机分为训练集(128例)和验证集(56例),并根据病理结果分为有颈部淋巴结转移者81例(转移组)和无颈部淋巴结转移者103例(无转移组)。采用单因素和多因素Logistic回归分析筛选临床资料中预测颈部淋巴结转移的影响因素;提取结节的超声影像组学特征,并采用最小绝对收缩和选择算法降维筛选有统计学意义的特征,采用支持向量机分类器分别构建基于超声影像组学的预测模型、超声影像组学联合临床特征的预测模型。绘制受试者工作特征(ROC)曲线分析各模型诊断PTMC患者颈部淋巴结转移的效能。结果单因素及多因素Logistic回归分析显示,年龄、包膜浸润均为PTMC患者淋巴结转移的独立预测因素(OR=3.131、3.245,均P<0.05)。共筛选出11个系数非零的超声影像组学特征。分别建立基于超声影像组学、超声影像组学联合临床特征的预测模型,ROC曲线分析显示,超声影像组学模型在训练集和验证集中诊断PTMC患者颈部淋巴结转移的曲线下面积(AUC)分别为0.78(95%可信区间:0.74~0.82)和0.72(95%可信区间:0.68~0.75);联合模型在训练集和验证集中诊断PTMC患者颈部淋巴结转移的AUC分别为0.87(95%可信区间:0.83~0.90)和0.81(95%可信区间:0.78~0.83),联合模型在训练集和验证集的AUC均高于超声影像组学模型(P=0.043、0.036)。结论基于超声影像组学联合临床特征的预测模型在PTMC患者颈部淋巴结转移的诊断中有一定的临床价值。

Objective To establish the predictive model based on ultrasound imageomics combined with clinical features,and to explore the clinical value of the model in diagnosing cervical lymph node metastasis in patients with papillary thyroid microcarcinoma(PTMC).Methods A total of 184 postoperative PTMC patients confirmed by pathology were randomly divided into a training set(n=128)and a validation set(n=56)in a 7∶3 ratio.According to the pathological results,the patients were classified into those with cervical lymph node metastasis(metastasis group,n=81)and those without cervical lymph node metastasis(non-metastasis group,n=103).Univariate and multivariate Logistic regression analysis were performed to identify influencing factors for predicting cervical lymph node metastasis in PTMC among clinical data.Ultrasound imageomics features of nodules were extracted,and statistically significant features were selected by the least absolute shrinkage and selection operator algorithm.Support vector machine(SVM)classifiers was used to construct predictive models based on ultrasound imageomics alone and combined with clinical features.Receiver operating characteristic curve(ROC)was drawn to analyze the performance of each model in diagnosing cervical lymph node metastasis in PTMC.Results Univariate and multivariate Logistic regression analysis showed that the age and capsule infiltration were independent predictors of lymph node metastasis in PTMC patients(OR=3.131,3.245,both P<0.05).A total of 11 ultrasound imageomics features with non-zero coefficients were selected.The predictive models based on ultrasound imageomics alone and combined with clinical features were established.ROC curve analysis showed that the area under the curve(AUC)of the ultrasound imaging model in the training set and validation set for diagnosing of cervical lymph node metastasis in PTMC patients was 0.78(95%confidence interval:0.74~0.82)and 0.72(95%confidence interval:0.68~0.75).The AUC of the combined model for diagnosing of cervical lymph node metastasis in PTMC patients in the training set and validation set were 0.87(95%confidence interval:0.83~0.90)and 0.81(95%confidence interval:0.78~0.83),the AUC of the combined model in the training set and validation set was higher than those of the ultrasound imaging model(P=0.043,0.036).Conclusion Ultrasound imageomics-based predictive model have certain clinical value in the diagnosis of cervical lymph node metastasis in PTMC patients.