Page 45 - 南京医科大学自然版

第45卷第2期 南京医科大学学报（自然科学版）
2025年2月 Journal of Nanjing Medical University（Natural Sciences） ·185 ·

·临床研究·

术前多区域多参数乳腺MRI影像组学对N2⁃3期腋窝淋巴结的

预测价值

林佳璐，陈佳雪，娄鉴娟，邹启桂，王思奇，蒋燕妮 *

南京医科大学第一附属医院放射科，江苏 南京 210029

［摘 要］ 目的：基于瘤内及瘤周多参数磁共振成像（magnetic resonance imaging，MRI）影像组学构建术前预测浸润性乳腺癌
N2⁃3期腋窝淋巴结（axillary lymph node，ALN）的列线图并验证模型效能。方法：回顾性分析2018年1月—2019年12月接受规
范治疗前乳腺MRI检查的320例（训练集224例，验证集96例）浸润性乳腺癌患者资料。根据病理报告中腋窝转移淋巴结数量
将患者分为N0⁃1期组（转移<4枚）和N2⁃3期组（转移≥4枚）。从5个MRI序列的瘤内、瘤周及瘤内+瘤周3个感兴趣区（region
of interest，ROI）中分别提取并筛选与ALN分期相关的影像组学特征，通过随机森林（random forest，RF）分类器构建3个影像组
学模型（瘤内模型、瘤周模型及瘤内+瘤周模型），计算各自的影像组学评分。采用单因素、多因素Logistic回归分析筛选临床独
立预测因子，构建临床模型。整合最优影像组学模型和临床模型中的特征构建联合模型，并可视化为列线图。通过受试者工
作特征（receiver operating characteristic，ROC）曲线、校准曲线和决策曲线分析（decision curve analysis，DCA）评估并比较各模型
的预测效能和临床实用价值。结果：在单纯影像组学模型中瘤内+瘤周模型的表现最优，训练集和验证集中的曲线下面积
（area under the curve，AUC）分别为0.911和 0.858。整合了瘤内+瘤周组学特征和临床特征（瘤周水肿和病灶强化形态）的列线
图在所有模型中具有最佳的N2⁃3期ALN预测效能，训练集和验证集的AUC分别为0.923和0.892。校准曲线显示列线图的预
测值与实际观测值一致性良好。DCA显示列线图临床效用较高。结论：联合多区域多参数MRI影像组学特征和临床特征构
建的列线图对浸润癌乳腺癌N2⁃3期腋窝淋巴结的个性化预测有较高价值。
［关键词］ 乳腺癌；N2⁃3期；腋窝淋巴结；磁共振成像；影像组学
［中图分类号］ R445.2 ［文献标志码］ A ［文章编号］ 1007⁃4368（2025）02⁃185⁃11
doi：10.7655/NYDXBNSN241100

Predictive value of preoperative multiregional multiparametric breast MRI radiomics for
N2⁃3 stage axillary lymph nodes

LIN Jialu，CHEN Jiaxue，LOU Jianjuan，ZOU Qigui，WANG Siqi，JIANG Yanni *
Department of Radiology，the First Affiliated Hospital of Nanjing Medical University，Nanjing 210029，China

［Abstract］ Objective：To develop and validate a radiomics nomogram based on intratumoral and peritumoral multiparametric
magnetic resonance imaging（MRI）for preoperative prediction of N2⁃3 stage axillary lymph node（ALN）in invasive breast cancer.
Methods：We retrospectively analyzed 320 invasive breast cancer patients（224 in the training set and 96 in the validation set）who
underwent preoperative standardized breast MRI from January 2018 to December 2019. Based on the number of axillary metastatic
lymph nodes in the pathological reports，patients were divided into N0⁃1 group（fewer than 4 metastatic nodes）and N2⁃3 group（4 or
more metastatic nodes）. Radiomics features associated with ALN stage were selected from three regions of interest（ROI）across five
MRI sequences：intratumor，peritumor，and intratumor+peritumor. Three radiomics models（intratumoral model，peritumoral model，
and intratumoral + peritumoral model）were constructed using a random forest（RF）classifier，and radiomics scores for each model
were calculated. Univariate and multivariate logistic regression were performed to identify clinical parameters and then constructed a
clinical model. Features from the optimal radiomics model and clinical model were integrated to build a combined model and visualized
as a nomogram. The predictive performance and clinical utility of each model were evaluated and compared using receiver operating

［基金项目］ 国家自然科学基金（NSFC61771249）；南京医科大学临床提升项目（2022NMUS0203）
通信作者（Corresponding author），E⁃mail：jyn_njmu@163.com（ORCID：0000⁃0003⁃1629⁃6777）
∗