基于常规超声的随机森林模型预测临床淋巴结阴性IBC前哨淋巴结转移  

Random Forest Model Based on Conventional Ultrasound Features for Predicting Sentinel Lymph Node Metastasis in Clinically Node-Negative Invasive Breast Cancer

作　　者：郑蒯锋 陆钊颖 庞锦宜 林嘉炜 周广新[2] 刘晓真 Zheng Kuaifeng;Lu Zhaoying;Pang Jinyi;Lin Jiawei;Zhou Guangxin;Liu Xiaozhen(Guangdong Medical University,Zhanjiang,Guangdong 524023,China;Department of Ultrasound,Zhongshan People's Hospital,Zhongshan,Guangdong 528403,China;Guangdong Medical University Zhongshan People's Hospital,Zhongshan,Guangdong 528403,China)

机构地区：[1]广东医科大学,广东省湛江市524023 [2]中山市人民医院超声影像科,广东省中山市528403 [3]广东医科大学中山市人民医院,广东省中山市528403

出　　处：《中国超声医学杂志》2025年第3期275-278,共4页Chinese Journal of Ultrasound in Medicine

摘　　要：目的 探讨基于临床淋巴结阴性浸润性乳腺癌(IBC)超声特征的随机森林模型预测前哨淋巴结转移的临床价值。方法 回顾性分析183例术后确诊为浸润性乳腺癌患者的临床以及超声资料,根据术后病理有无淋巴结转移分为转移组(LNM组)44例、未转移组(NLNM组)139例,按照7∶3随机分为训练组128例、测试组55例。比较两组的临床以及超声资料的差异性,应用Logistic回归分析筛选出浸润性乳腺癌淋巴结转移的独立危险因素,并使用随机森林构建预测模型。结果 多因素Logistic分析最大径、高回声晕环及血流信号可以作为预测转移组与未转移组的独立危险因素,随机森林算法构建的预测模型的曲线下面积训练组为0.915,测试组为0.889。结论 基于常规超声的基本特征,利用随机森林算法构建的模型可以有效预测浸润性乳腺癌患者的前哨淋巴结转移情况。Objective To explore the clinical value of a random forest model based on ultrasonic characteristics of clinically node-negative invasive breast cancer(IBC)in predicting sentinel lymph node metastasis.Methods A retro-spective analysis of 183 patients diagnosed with invasive breast cancer after surgery was conducted.Clinical and ultra-sound data were reviewed,and patients were divided into two groups based on the presence of lymph node metastasis in postoperative pathology:the lymph node metastasis(LNM)group with 44 cases and the non-lymph node metastasis(NLNM)group with 139 cases.Patients were randomly assigned to a training group of 128 cases and a test group of 55 cases in a 7:3 ratio.Differences in clinical and ultrasound data between the two groups were compared,and Logis-tic regression analysis was applied to identify independent risk factors for lymph node metastasis in invasive breast cancer.A predictive model was constructed using a random forest algorithm.Results Multivariate Logistic analysis identified the maximum diameter,hyperechoic halo and blood flow signals as independent risk factors for predicting the metastasis group versus the non-metastasis group.The area under the curve(AUC)of the predictive model con-structed by the random forest was 0.915 for the training group and 0.889 for the test group.Conclusions A model constructed using the random forest algorithm based on basic characteristics from routine ultrasound can effectively predict sentinel lymph node metastasis in patients with invasive breast cancer.

关 键 词：超声特征 随机森林 浸润性乳腺癌 前哨淋巴结 

分 类 号：R737.9[医药卫生—肿瘤] R445.1[医药卫生—临床医学]