多参数MRI影像组学和机器学习模型可较准确预测三阴性乳腺癌  

Multiparameter MRI radiomcs and machine learning-based models can accurately predict triple-negative breast cancer

作　　者：戚轩 王武陵 杨宏楷[1] 程卫群 翟承凤 杨馨 段绍峰 何永胜[1] QI Xuan;WANG Wuling;YANG Hongkai;CHENG Weiqun;ZHAI Chengfeng;YANG Xin;DUAN Shaofeng;HE Yongsheng(Department of Radiology,Ma'anshan People's Hospital,Ma'anshan 243000,China;The Fifth Clinical Medical College of Anhui Medical University,Ma'anshan Clinical College,Anhui Medical University,Ma'anshan 243000,China)

机构地区：[1]马鞍山市人民医院影像科,安徽马鞍山243000 [2]安徽医科大学第五临床医学院,安徽医科大学马鞍山临床学院,安徽马鞍山243000

出　　处：《分子影像学杂志》2025年第1期82-90,共9页Journal of Molecular Imaging

基　　金：马鞍山市卫科技计划项目(YL-2024-07);马鞍山市卫生健康科重点科研项目(MASWJ2022a001);安徽医科大学校科研基金项目(2023xkj122)。

摘　　要：目的通过对多参数磁共振成像数据进行影像组学特征提取,联合临床特征建立预测模型,寻找对三阴性乳腺癌(TNBC)预测价值最高的机器学习模型。方法收集175例乳腺癌患者,包括40例TNBC和135例非TNBC患者,按7∶3分为训练集(n=123)和验证集(n=52),使用不同的机器学习算法建立多参数的预测模型,并与临床特征联合建模,通过ROC曲线评价不同模型的预测效能。结果在训练集和验证集中,病灶边界、WHO分级及T2WI信号在TNBC和非TNBC中的差异有统计学意义(P<0.05),基于rbf_SVM建立的Model-T2WI、Model-DWI、Model-DCE_(phase2)、Model-DCE_(phase7)、T2WI+DWI、DCE_(Phase7)+T2WI、DCE_(Phase7)+DWI、DCE_(Phase7)+T2WI+DWI、DCE_(Phase7)+T2WI+DWI+Clinic的9个模型中,DCE_(Phase7)+T2WI+DWI+Clinic的影像组学建立的预测模型效能最高,在训练集和验证集中的曲线下面积分别为0.992、0.936。结论基于多参数磁共振成像的影像组学模型能较准确地预测TNBC,有助于TNBC的临床诊疗管理。Objective To establish a predictive model by extracting radiomic features from multi-parametric MRI data and combining them with clinical characteristics,and identify the machine learning model with the highest predictive value for triple-negative breast cancer(TNBC).Methods A total of 175 breast cancer patients,including 40 cases of TNBC and 135 cases of non-triple negative breast cancer(NTNBC),were collected and divided into training set(n=123)and validation set(n=52)according to 7:3.Multiparameter predictive models were developed using various machine learning algorithms and combined with clinical features for joint modeling.The predictive performance of different models was assessed using ROC curves.Results In the training and validation sets,Boundary,WHO classification and T2WI signals of lesions were statistically different in TNBC and NTNBC(P<0.05),among the nine models established using rbf_SVM,including Model-T2WI,Model-DWI,Model-DCE_(Phase2),Model-DCE_(Phase7),Model-T2WI+DWI,Model-DCE_(Phase7)+T2WI,Model-DCE_(Phase7)+T2WI+DWI,and Model-DCE_(Phase7)+T2WI+DWI+Clinic,the radiomics-based predictive model of Model-DCEPhase7+T2WI+DWI+Clinic demonstrated the highest performance,with areas under the curve(AUC)of 0.992 and 0.936 in the training and validation sets,respectively.Conclusion The radiomics model based on multi-parametric MRI can accurately predict TNBC,contributing to the clinical diagnosis and treatment management of TNBC.

关 键 词：乳腺癌 三阴性乳腺癌 影像组学 多参数磁共振 

分 类 号：R73[医药卫生—肿瘤]