摘要
目的本研究旨在构建一个基于临床和影像学特征的极端梯度提升(extreme gradient boosting,XGBoost)模型,以鉴别乳腺非肿块病变的良恶性。材料与方法收集2018年1月至2024年7月2个机构,2种乳腺X线设备检查的有病理结果的首诊乳腺非肿块病变480个。患者被分为建模组[n=310,数字乳腺X线摄影(digital mammography,DM)检查]、内部验证组(n=108,DM检查),和外部验证组[n=62,数字乳腺体层合成摄影(digital breast tomosynthesis,DBT)检查]。记录患者术前乳腺X线(DM或DBT),MRI以及临床特征。采用XGBoost算法和多因素逻辑回归分析,分别构建XGBoost模型和逻辑回归(logistic regression,LR)模型。使用受试者工作特征(receiver operating characteristic,ROC)曲线评估模型的诊断效能。结果在建模组中,患者以7∶3随机分为训练集(n=217)和测试集(n=93)。训练集、测试集、训练集的内部验证组及训练集的外部验证组中,恶性非肿块病灶分别为159(73%)、58(62%)、73(68%)和43(69%)。XGBoost模型的诊断效能明显优于LR模型,在独立的训练集、测试集、训练集的内部验证组及训练集的外部验证组中均表现出良好的诊断效能,曲线下面积(area under the curve,AUC)在0.884~0.913之间。XGBoost模型在四个队列中也表现出良好的校准能力和临床净获益。结论XGBoost模型能够准确鉴别乳腺非肿块病变的良恶性,具有推广应用的潜力。
Objective:To develop an extreme gradient boosting(XGBoost)model based on clinical and imaging features to differentiate between benign and malignant non-mass breast lesions.Materials and Methods:Data were collected from January 2018 to July 2024 from two institutions,focusing on 480 non-mass breast lesions with pathological results obtained from two types of mammography equipment.Patients were categorized into a modeling group[n=310,digital mammography(DM)examination],an internal validation group(n=108,DM examination),and an external validation group[n=62,digital breast tomosynthesis(DBT)examination].Preoperative breast X-ray(DM or DBT),MRI,and clinical characteristics were recorded.The XGBoost algorithm and multivariate logistic regression(LR)analysis were employed to develop the XGBoost and LR models,respectively.Diagnostic performance was assessed using receiver operating characteristic(ROC)curves.Results:In the modeling group,patients were randomly split in a 7∶3 ratio into a training set(n=217)and a test set(n=93).The proportion of malignant non-mass lesions in the training set,test set,internal validation group of the training set,and external validation group of the training set,were 159(73%),58(62%),73(68%)and 43(69%),respectively.The XGBoost model outperformed the LR model in diagnostic accuracy,demonstrating superior performance across the independent training,test,and internal,external validation sets of the training set,with area under the curve(AUC)ranging from 0.884 to 0.913.Additionally,the XGBoost model exhibited good calibration and clinical net benefit in all four cohorts.Conclusions:The XGBoost model accurately differentiates between benign and malignant non-mass breast lesions,indicating its potential for widespread clinical application.
作者
杨文
杨蔚
周晓平
杨妍
张宁妹
尹清云
张朝林
刘召弟
YANG Wen;YANG Wei;ZHOU Xiaoping;YANG Yan;ZHANG Ningmei;YIN Qingyun;ZHANG Chaolin;LIU Zhaodi(The First School of Clinical Medicine,Ningxia Medical University,Yinchuan 750004,China;Department of Radiology,General Hospital of Ningxia Medical University,Yinchuan 750004,China;Information Technology Center,32752 Troop,Xiangyang 441000,China;Department of Pathology,General Hospital of Ningxia Medical University,Yinchuan 750004,China;Department of Medical Oncology,General Hospital of Ningxia Medical University,Yinchuan 750004,China;Department of Oncology Surgery,General Hospital of Ningxia Medical University,Yinchuan 750004,China;Department of Radiology,First People's Hospital of Shizuishan City,Shizuishan 753200,China)
出处
《磁共振成像》
北大核心
2025年第1期118-126,145,共10页
Chinese Journal of Magnetic Resonance Imaging
基金
宁夏回族自治区重点研发计划项目(编号:2022BEG03166)
宁夏回族自治区自然科学基金项目(编号:2024AAC02070)。
关键词
非肿块强化
乳腺癌
极端梯度提升
机器学习
磁共振成像
乳腺X线摄影
non-mass enhancement
breast cancer
extreme gradient boosting
machine learning
magnetic resonance imaging
mammography
