目的 探讨多模态影像组学列线图术前预测乳腺浸润性导管癌腋窝淋巴结(axillary lymph node,ALN)转移的价值.材料与方法 回顾性分析2019年1月至2023年6月在我院经手术病理证实为乳腺浸润性导管癌的224例患者的临床及影像资料.首先,选取T2WI图像和动态对比增强MRI(dynamic contrast-enhanced MRI,DCE-MRI)第二期图像的病灶最大层面及同一病灶的钼靶(mammography,MG)头尾位、内外斜位图像勾画感兴趣区,并且提取病灶感兴趣区特征,按照7∶3比例将样本随机分为训练集156例和测试集68例,通过最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)回归进行特征降维筛选,选择5种机器学习分类器[支持向量机(support vector machine,SVM)、K近邻(K nearest neighbors,KNN)、极端梯度提升决策树(extreme gradient boosting,XGBoost)、逻辑回归(logistic regression,LR)、随机森林(randomforest,RF)]构建多模态影像组学模型并选择预测性能最佳分类器建立MRI、MG影像组学模型.通过单-多因素logistic回归筛选临床高危因素构建临床模型.最终选择影像组学评分联合临床高危因素构建影像组学列线图模型.采用受试者工作特征(receiver operating characteristic,ROC)曲线及曲线下面积(area under the curve,AUC)评价模型预测乳腺癌患者ALN状态的性能,利用校准曲线评价模型的拟合能力,决策曲线评估预测模型的临床实用性.结果 最终得到14个最佳影像组学特征.在测试集中5种机器学习分类器AUC值范围为0.764～0.864,其中SVM的AUC值最高(0.864).淋巴结触诊(P<0.001)及MRI_ALN(P=0.005)是评估ALN是否转移的独立危险因素.列线图模型训练集AUC、敏感度、特异度及准确度分别为0.941、90.7%、88.9%、88.5%;测试集分别为0.926、84.4%、86.1%、85.3%.结论 列线图模型性能最佳,在术前预测ALN状态具有重要的价值,可以协助临床制订科学有效的诊疗方案.
Value of multimodal radiomics nomogram in predicting axillary lymph node metastasis in invasive ductal carcinoma of the breast before surgery
Objective:To investigate the value of multimodal radiomics nomogram in predicting axillary lymph node(ALN)metastasis in invasive ductal carcinoma of the breast before surgery.Materials and Methods:The clinical and imaging data of 224 patients with invasive ductal carcinoma of the breast confirmed by surgical pathology in our hospital from January 2019 to June 2023 were retrospectively collected.Firstly,the maximum level of the lesion of the T2WI image and the second phase of dynamic contrast-enhanced MRI(DCE-MRI)and the mammography(MG)of the same lesion were selected to delineate the region of interest,and the characteristics of the lesion area of interest were extracted.According to the ratio of 7∶3,the samples were randomly divided into 156 cases in the training set and 68 cases in the test set,and the feature dimensionality reduction screening was carried out by least absolute shrinkage and selection operator(LASSO)regression,5 kinds of machine learning classifiers[support vector machine(SVM)、K nearest neighbors(KNN)、extreme gradient boosting(XGBoost)、logistic regression(LR)、randomforest(RF)]were selected to build a multimodal radiomics model,and the classifier with the best prediction performance was selected to establish MRI and mammography models.Univariate logistic regression was used to screen clinical high-risk factors and construct a clinical model.Finally,Radiomics score combined with clinical high-risk factors was selected to construct an electromics nomogram model.The receiver operating characteristic(ROC)curve and area under the curve(AUC)were used to evaluate the efficacy of the model in predicting the ALN status of breast cancer patients,and the clinical practicability of the prediction model was evaluated by using the fitting ability of the calibration curve to evaluate the decision curve.Results:Finally,14 optimal radiomics features were obtained.The AUC value of the five machine learning classifiers in the test set ranged from 0.764-0.864,and the AUC value of SVM was the highest(0.864).Lymph node palpation(P<0.001)and MRI_ALN(P=0.005)were independent risk factors for ALN metastasis.The AUC,sensitivity,specificity and accuracy of the nomogram model training set were 0.941,90.7%,88.9%and 88.5%,respectively.The test sets were 0.926,84.4%,86.1%,and 85.3%,respectively.Conclusions:The nomogram model has important value in predicting ALN status before surgery,and can assist in the formulation of scientific and effective clinical diagnosis and treatment plans.
breast cancerinvasive ductal carcinomaaxillary lymph nodesradiomicsnomogramsmammographymagnetic resonance imaging
国家科技期刊平台
NSTL
万方数据
维普
