基于数字乳腺断层摄影的瘤内和瘤周影像组学对乳腺癌淋巴血管浸润状态的术前预测  

Preoperative prediction of lymphovascular invasion in breast cancer with digital breast tomosynthesis-based intratumoral and peritumoral radiomics

作　　者：张素鑫 李海燕[1] 郑逸群 陈文青 何生[2] 杨彩仙[3] 梁刚 李健丁[2] 姜增誉[2] ZHANG Suxin;LI Haiyan;ZHENG Yiqun;CHEN Wenqing;HE Sheng;YANG Caixian;LIANG Gang;LI Jianding;JIANG Zengyu(School of Medical Imaging,Shanxi Medical University,Taiyuan 030001,China;Department of Radiology,the First Hospital of Shanxi Medical University,Taiyuan 030001,China;Department of Radiology,Shanxi Provincial People’s Hospital,Taiyuan 030012,China;Department of Pathology,the First Hospital of Shanxi Medical University,Taiyuan 030001,China)

机构地区：[1]山西医科大学医学影像学院,山西太原030001 [2]山西医科大学第一医院影像科,山西太原030001 [3]山西省人民医院放射科,山西太原030012 [4]山西医科大学第一医院病理科,山西太原030001

出　　处：《实用放射学杂志》2025年第1期46-51,共6页Journal of Practical Radiology

基　　金：山西省科技兴医四个“一批”创新项目(2023XM031)。

摘　　要：目的基于数字乳腺断层摄影(DBT)瘤内及瘤周区域影像组学诺模图预测乳腺癌患者淋巴血管浸润(LVI)状态。方法回顾性选取2个机构的192例乳腺癌患者,机构1的162例患者用于训练(n=113)和测试(n=49),机构2的30例患者用于外部验证。在DBT图像中提取并选择基于瘤内及瘤周1 mm区域的影像组学特征,利用不同机器学习算法分别建立模型_(瘤内)、模型_(瘤周)及模型_(瘤内+瘤周),对临床资料进行单因素和多因素logistic分析确定独立危险因素并构建模型_(临床影像),使用受试者工作特征(ROC)曲线对模型效能进行评估,联合诊断效能最优的影像组学特征与筛选出的临床影像特征构建综合模型_(临床-影像组学),并绘制诺模图。结果模型_(瘤内+瘤周)为最优影像组学模型。最大肿瘤直径[比值比(OR)=1.486,P=0.014]、可疑恶性钙化(OR=2.898,P=0.015)及腋窝淋巴结(ALN)转移(OR=3.615,P<0.001)为LVI阳性的独立危险因素。综合模型_(临床-影像组学)在训练集、测试集及外部验证集的曲线下面积(AUC)分别为0.889、0.916、0.862,均高于模型_(瘤内+瘤周)(0.858、0.849、0.844)及模型_(临床影像)(0.743、0.759、0.732)。结论从影像组学和临床影像学特征得出的预测诺模图可以相对准确识别乳腺癌中未来LVI的发生率,证明其辅助临床医师设计个体化治疗方案的潜力。Objective To predict the lymphovascular invasion(LVI)status of breast cancer patients based on digital breast tomosynthesis(DBT)intratumoral and peritumoral radiomics nomogram.Methods A total of 192 breast cancer patients from 2 institutions were retrospectively selected,in which institution 1 was used for train(n=113)and test(n=49),while institution 2 was used for external validation(n=30).Radiomics features were extracted and selected based on intratumoral and peritumoral 1 mm regions from DBT images.Different machine learning algorithms were used to construct intratumoral,peritumoral,and combined intratumoral and peritumoral models,respectively.Patient clinical data were analyzed by both univariate and multivariate logistic regression analyses to identify independent risk factors for the clinical imaging model.The performance of the models was evaluated using the receiver operating characteristic(ROC)curve.The radiomics features with the optimal diagnostic performance and the selected clinical imaging features were combined to construct a comprehensive clinical-radiomics model,and a nomogram was drawn.Results The combined intratumoral and peritumoral model was the optimal radiomics model.Maximum tumor diameter[odds ratio(OR)=1.486,P=0.014],suspicious malignant calcifications(OR=2.898,P=0.015),and axillary lymph node(ALN)metastasis(OR=3.615,P<0.001)were independent risk factors for LVI positive.Furthermore,the area under the curve(AUC)of the comprehensive clinical-radiomics model in the training set,test set and external validation set was 0.889,0.916,and 0.862,respectively,which was higher than those of the combined intratumoral and peritumoral model(0.858,0.849,0.844)and the clinical imaging model(0.743,0.759,0.732).Conclusion The predictive nomogram,derived from both radiomics and clinical imaging features,is relatively accurate in identifying future LVI occurrence in breast cancer,demonstrating its potential as an assistive tool for clinicians to devise individualized treatment regimes.

关 键 词：乳腺癌 影像组学 数字乳腺断层摄影 淋巴血管浸润 

分 类 号：R737.9[医药卫生—肿瘤] R445[医药卫生—临床医学] R543