基于DCE-MRI瘤内及瘤周影像组学特征的机器学习模型预测乳腺癌腋窝淋巴结转移的价值

The Value of Machine Learning Models for Predicting Axillary Lymph Node Metastasis in Breast Cancer Based on Intratumoral and Peritumoral Radiomics Features of DCE-MRI

目的:探讨基于动态增强磁共振成像(DCE-MRI)瘤内及瘤周影像组学特征的机器学习模型预测乳腺癌腋窝淋巴结转移的价值。方法:回顾性收集2018年6月一2022年8月经病理证实为乳腺癌且有相应腋窝淋巴结状态的患者病例资料215例,其中腋窝淋巴结转移阳性的患者117例,阴性98例。以7:3的比例将所有患者随机分为训练集与验证集。在训练集中使用皮尔森相关系数(PCC),最大相关最小允余(mRMR)、最小绝对收缩和选择算子(LASSO)对影像组学数据进行降维,采用逻辑回归(LR)、随机森林(RF)、支持向量机(SVM)、K近邻(KNN)、决策树(DT)、线性判别(LDA)6种方法建立影像组学预测模型。通过受试者工作特征曲线下面积(AUC)评价模型的效能。结果:基于瘤内联合瘤周影像组学构建的预测模型,在训练集及验证集中使用LR、RF、SVM、KNN、DT、LDA作为分类器时的AUC分别为(训练集0.811、0.882、0.816、0.773、0.870、0.829,验证集0.787、0.824、0.815、0.711、0.727、0.799),RF模型的预测效能最佳,且高于RF构建的单独瘤内模型(训练集AUC=0.837,验证集AUC=0.599)和瘤周模型(训练集AUC=0.815,验证集AUC=0.764)效能。结论:基于DCE-MRI瘤内联合瘤周影像组学特征构建的机器学习模型对预测乳腺癌患者腋窝淋巴结转移有着较高的价值,其中RF模型的预测能力最佳。

Purpose:To investigate the value of a machine learning model for predicting axillary lymph node metastasis in breast cancer based on the intratumoral and peritumoral radiomics features of dynamic contrast enhancement magnetic resonance imaging(DCE-MRI).Methods:A total of 215 patients with pathologically confirmed breast cancer with corresponding axillary lymph node status were retrospectively collected from June 2018 to August 2022,including 117 patients with positive axillary lymph node metastasis and 98 negative cases.All 215 patients were divided into training set and validation set with a ratio of 7:3 by complete randomization method.Pearson correlation coefficients(PCC),max-relevance and min-redundancy(mRMR),least absolute shrinkage and selection operator(LASSO)were used for dimensional reduction of radiomics data in the training set.Logistic regression(LR),random forest(RF),support vector machine(SVM),K-nearest neighbor(KNN),decision tree(DT)and linear discriminant analysis(LDA)were used to establish the radiomics prediction model.Area under the ROC curve(AUC)was used to evaluate the efficacy of the model.Results:Based on the prediction model constructed by intratumoral combined peritumoral radiomics,the AUCs of classifiers of LR,RF,SVM,KNN,DT and LDA in the training set and validation set were as follows:0.811,0.882,0.816,0.773,0.870,0.829 in the training set,and 0.787,0.824,0.815,0.711,0.727,0.799 in the validation set,respectively.The RF model had the best predictive performance.The intratumoral model(the AUC of training set:0.837;the AUC of validation set:0.599)and peritumoral model(the AUC of training set:0.815;the AUC of validation set:0.764)alone was with lower diagnostic efficacy than the intratumoral combined peritumoral model.Conclusion:The machine learning model based on intratumoral combined peritumoral radiomics features of DCE-MRI has good values in predicting axillary lymph node metastasis of breast cancer,and the RF model has the best predictive ability.