基于乳腺动态对比增强MRI深度学习与影像组学联合临床特征预测乳腺浸润性导管癌脉管浸润状态  

Prediction of Vascular Invasion Status of Invasive Ductal Carcinoma of the Breast Using Deep Learning and Radiomics Based on DCE-MRI

作　　者：谢汉民 黄煌[1] 杨朝湘[1] 张若仙 黄育斌[1] XIE Hanmin;HUANG Huang;YANG Chaoxiang(Department of Radiology,Guangdong Women and Children Hospital,Guangzhou,Guangdong Province 511400,P.R.China)

机构地区：[1]广东省妇幼保健院放射科,广州511400

出　　处：《临床放射学杂志》2025年第5期832-838,共7页Journal of Clinical Radiology

摘　　要：目的探讨基于动态对比增强磁共振成像(DCE-MRI)深度迁移学习(DTL)特征、传统影像组学特征以及临床特征构建的联合模型预测乳腺浸润性导管癌脉管浸润(LVI)状态的应用价值。方法回顾性分析经病理证实的185例乳腺浸润性导管癌患者,按照脉管内有无癌栓分为LVI(+)(82例)和LVI(-)(103例)。按照8∶2随机划分为训练集[148例,其中62例LVI(+)、86例LVI(-)]和验证集[37例,其中20例LVI(+)、17例LVI(-)]。采用多因素Logistic回归分析乳腺浸润性导管癌LVI的独立危险因素。使用ITK-SHAP 3.8.0软件对DCE-MRI第三期增强图像进行肿瘤感兴趣区(ROI)勾画,采用Pyradiomics和Python分别提取DCE-MRI影像中病灶的影像组学特征和DTL特征,用于构建DTL特征数据集、影像组学特征数据集和DTL特征与影像组学特征融合数据集,分别以统计检验、Pearson分析、最小冗余最大相关性(mRMR)和最小绝对收缩和选择算子(LASSO)回归对各特征数据集进行特征筛选降维,以其构建传统影像组学模型、DTL模型以及二者融合模型(DLR模型),并筛选出最佳影像组学模型,基于最佳影像组学模型评分与临床独立危险因素构建联合模型,并计算曲线下面积(AUC)、准确率、敏感度、特异度值指标评估各模型效能,绘制校准曲线评估模型拟合度,以决策曲线分析评估模型的应用价值。结果经单因素及多因素Logistic回归分析,年龄(OR=0.971,95%CI:0.955~0.988,P=0.005)及前哨淋巴结转移(OR=6.363,95%CI:3.209~12.616,P=0.000)为乳腺癌LVI的独立危险因素。DLR模型为最佳影像组学模型,联合模型效能均高于其他特征模型,在训练集中AUC为0.990,验证集中AUC为0.824。联合模型的拟合度更高,临床获益度更大。结论基于DCE-MRI图像的DTL特征、影像组学特征以及临床特征构建的联合模型能较好地预测乳腺浸润性导管癌LVI状态。Objective To explore the application value of a combined model based on DCE-MRI deep transfer learning(DTL)features,traditional radiomics(Rad)features,and clinical features in predicting vascular invasion(LVI)status of breast invasive ductal carcinoma.Methods A total of 185 patients with breast invasive ductal carcinoma confirmed by pathology were retrospectively analyzed.According to the presence or absence of tumor thrombus in the blood vessels,they were divided into LVI(+)(82 cases)and LVI(-)(103 cases).Patients were randomly divided into a training set[148 cases,of which 62 cases had LVI(+)and 86 cases had LVI(-)]and a validation set[37 cases,of which 20 cases had LVI(+)and 17 cases had LVI(-)]at a ratio of 8∶2.Multivariate Logistic regression was used to analyze the independent risk factors for vascular invasion of breast invasive ductal carcinoma.ITK-SNAP 3.8.0 software was used to delineate the region of interest(ROI)of the tumor.Pyradiomics and Python were used to extract Rad and DTL features of lesions in DCE-MRI images.For the construction of DTL feature dataset,Rad feature dataset,and DTL and Rad feature fusion dataset,statistical tests,Pearson analysis,minimum redundancy maximum relevance(mRMR),and least absolute shrinkage and selection operator(LASSO)regression were used to select features and reduce the dimensionality of each feature dataset.Based on the best radiomics model score and clinical independent risk factors,a combined model was constructed.The area under the curve(AUC),accuracy,sensitivity,and specificity were calculated to evaluate the performance of each model.Calibration curves were drawn to assess the fit of the model,and decision curve analysis was used to evaluate the clinical application value of the model.Results Univariate and multivariate logistic regression analysis showed that age(OR=0.971,95%CI:0.955-0.988,P=0.005)and sentinel lymph node metastasis(OR=6.363,95%CI:3.209-12.616,P<0.001)were independent risk factors for vascular invasion of breast cancer.The DLR model was iden

关 键 词：磁共振成像 深度学习 脉管浸润 预测 浸润性导管癌 

分 类 号：R737.9[医药卫生—肿瘤]