基于超声图像影像组学分阶段预测乳腺癌腋窝淋巴结转移负荷  

Stage Prediction of Axillary Axillary Node Metastasis Status in Breast Cancer Based on Ultrasound Radiomics

作　　者：杨洁[1] 吴萌 冯晓丹 杜瑶 崔广和[1] 刘菲菲[1] Yang Jie;Wu Meng;Feng Xiaodan;Du Yao;Cui Guanghe;Liu Feifei(Department of Ultrasound,Binzhou Medical University Hospital,Binzhou,Shandong 256603,China)

机构地区：[1]滨州医学院附属医院超声医学科,山东省滨州市256603

出　　处：《中国超声医学杂志》2024年第11期1217-1220,共4页Chinese Journal of Ultrasound in Medicine

基　　金：山东省自然科学基金项目(No.ZR2023QH231)。

摘　　要：目的探讨基于超声图像影像组学特征的机器学习模型术前分阶段预测乳腺癌腋窝淋巴结转移负荷的临床价值。方法回顾性纳入332例早期乳腺癌患者,依据病理结果将淋巴结转移负荷分为无转移(N0)200例、低转移负荷[N+(1~2)]92例、高转移负荷[N+(≥3)]40例。基于乳腺癌原发灶超声图像提取并筛选影像组学特征,共筛选出7个关键影像组学特征构建腋窝淋巴结有无转移预测模型(模型1),6个关键影像组学特征构建腋窝淋巴结受累程度预测模型(模型2),两模型均基于随机森林算法构建。根据受试者工作特征曲线下面积(AUC)评估模型预测效能,并计算其准确度、灵敏度、特异度及F1值,应用决策曲线评估模型的临床实用性。结果预测腋窝淋巴结有无转移[N0与N+(≥1)]的模型1在训练集和测试集的AUC分别为0.79(95%CI:0.74~0.85)、0.77(95%CI:0.68~0.86),预测腋窝淋巴结受累程度[N+(1~2)与N+(≥3)]的模型2在训练集和测试集的AUC分别为0.86(95%CI:0.79~0.94)、0.83(95%CI:0.73~0.93)。决策曲线分析显示,两模型均具有较好的临床实用性。结论基于乳腺癌原发灶超声图像影像组学特征构建的机器学习模型能较准确地预测早期乳腺癌患者腋窝淋巴结转移负荷,有望在术前辅助临床医师制订个体化腋窝管理方案。Objective To investigate the clinical value of machine learning model based on ultrasonographic features to predict the axillary node metastatic load before surgery.Methods A total of 332 patients with early breast cancer were retrospectively included.According to the postoperative pathology,axillary node metastatic load was divided into 200 cases without metastasis(N0),92 cases with low metastatic load[N+(1-2)],and 40 cases with high metastatic load[N+(≥3)].Based on ultrasound images of the primary breast cancer,we extracted and screened the image omics features.A total of 7 key image omics features were screened to construct the axillary node metastasis prediction model(model 1),and 6 key image omics features were selected to construct the axillary node involvement degree prediction model(model 2).Both models were constructed based on random forest algorithm.The model performance was evaluated according to the area under receiver operating characteristic curve(AUC),and its accuracy,sensitivity,specificity and F1 values were calculated.The clinical practicability of the model was evaluated by the decision curve.Results The AUC values of model 1,which predicted axillary node metastasis[N0 and N+(≥1)]in the training set and test set were 0.79(95%CI:0.74-0.85)and 0.77(95%CI:0.68-0.86),respectively.The AUC values of Model 2 predicting axillary node involvement degree[N+(1-2)and N+(≥3)]in the training set and test set were 0.86(95%CI:0.79-0.94)and 0.83(95%CI:0.73-0.93),respectively.Decision curve analysis shows that both models had good clinical practicability.Conclusions The machine learning model built based on the image omics characteristics of primary breast cancer ultrasound images can predict the axillary lymph node metastatic load in early breast cancer patients accurately,which is expected to assist clinicians to develop individualized axillary management plan before surgery.

关 键 词：影像组学 超声 乳腺癌 腋窝淋巴结 

分 类 号：R445.1[医药卫生—影像医学与核医学] R737.9[医药卫生—诊断学]