基于乳腺X线图像影像组学列线图对乳腺癌腋窝淋巴结转移的预测价值

Predictive value of breast cancer axillary lymph node metastasis based on the radiomics nomogram of mammography

目的:探讨基于乳腺X线图像影像组学列线图对乳腺癌腋窝淋巴结(ALN)转移的预测价值。方法:回顾性分析188例乳腺癌患者的乳腺X线图像和临床资料,按照7:3的比例将患者随机分割为训练组(n=130)和验证组(n=58)。使用MaZda软件在乳腺X线图像内提取影像组学特征,应用方差选择法和最小绝对收缩与选择算子算法(LASSO)对提取的特征参数进行降维后建立影像组学标签,采用ROC曲线下面积(AUC)对训练组和验证组的影像组学标签的诊断效能进行评价;对临床病理特征进行单因素Logistic回归分析,联合影像组学标签和独立临床预测因子构建联合预测模型并绘制影像组学列线图,通过绘制校正曲线评估其标定,计算预测模型的AUC、敏感度、特异度。最后采用决策分析曲线评价列线图在不同风险阈值下的净获益情况。结果:从乳腺X线图像中提取了317个影像组学特征,利用LASSO算法筛选出14个价值较高的影像组学特征。由14个与乳腺癌ALN转移相关特征构建的影像组学标签分别达到了中等预测效果,训练组和验证组的AUC分别为0.760和0.742。肿瘤大小和影像组学标签构建的影像组学列线图有着较高的校准性能和预测性能,训练组和验证组的AUC分别为0.808和0.811。决策曲线显示影像组学列线图在5%~82%阈值范围内表现出良好的临床应用效能。结论:基于乳腺X线图像建立的影像组学列线图可以作为一种无创性的预测工具,帮助临床医生在术前确定乳腺癌患者的腋窝淋巴结状态。

Objective:To investigate the predictive value of breast cancer axillary lymph node(ALN) metastasis based on the radiomics nomogram of mammography.Methods:The mammography and clinical data of 188 patients with breast cancer were retrospectively analyzed.The patients were randomly divided into training cohort(n=130) and validation cohort(n=58) at a ratio of 7:3.The radiomics features of mammography were extracted by the Mazda software, and the radiomics signature was then constructed after reduction in dimension of the extracted characteristic parameters by variance selection and the least absolute shrinkage and selection operator algorithm(LASSO).The area under the ROC curve(AUC) was used to evaluate the diagnostic efficacy of the radiomics signature in training cohort and validation cohort.The clinicopathological features were analyzed by univariate logistic regression, and the joint predictive model was constructed by combining the radiomics signature and independent clinical predictors, and the radiomics nomogram was drawn.The calibration curve was used to evaluate the model and the AUC,the sensitivity and specificity of the nomogram were also calculated.Finally, decision curve analysis was conducted to evaluate the net benefits of radiomics nomogram at different threshold.Results:A total of 317 radiomics features were extracted from the mammography, and 14 most valuable features were selected by LASSO algorithm.The radiomics signatures, consisted of 14 features associated with ALN metastasis in breast cancer, achieved moderate predictive efficacy with AUC of 0.760 and 0.742 in training cohort and validation cohort, respectively.The radiomics nomogram, comprising tumor size and radiomics signatures, showed good calibration and predictive performance, with AUC of 0.808 and 0.811 in training cohort and validation cohort, respectively.The decision curve demonstrated the radiomics nomogram displayed good clinical utility in the range of 5% to 82% of the threshold.Conclusions:The radiomics nomogram based on the mammogram can be used as a non-invasive predictive tool to assist clinicians in determining ALN status in breast cancer preoperatively.