多模态影像组学列线图模型预测三阴性乳腺癌的价值  

Value of multimodal radiomics nomogram model in predicting triple-negative breast cancer

作　　者：朱芸 马宜传 杨丽 张舒妮 赵楠楠 杨静茹 王玲玲 甘浩然 谢宗玉 ZHU Yun;MA Yichuan;YANG Li;ZHANG Shuni;ZHAO Nannan;YANG Jingru;WANG Lingling;GAN Haoran;XIE Zongyu(Department of Radiology,The First Affiliated Hospital of Bengbu Medical University,Bengbu Anhui 233004;Department of Medical Imaging Diagnostics,Bengbu Medical University,Bengbu Anhui 233030,China;School of Clinical Medicine,Bengbu Medical University,Bengbu Anhui 233030,China)

机构地区：[1]蚌埠医科大学第一附属医院放射科,安徽蚌埠233004 [2]蚌埠医科大学医学影像诊断学教研室,安徽蚌埠233030 [3]蚌埠医科大学临床医学院,安徽蚌埠233030

出　　处：《蚌埠医学院学报》2024年第4期431-437,共7页Journal of Bengbu Medical College

基　　金：安徽省高校自然科学研究重点项目(2022AH051473,2023AH051947);蚌埠医学院自然科学研究重点项目(2021byzd118);蚌埠医学院临床研究专项(2022byflc008);安徽省大学生创新训练项目(S202310367044)。

摘　　要：目的:探讨基于钼靶(MG)双体位联合MRI双序列的多模态影像组学列线图模型在术前预测三阴性乳腺癌(TNBC)的价值。方法:分析经病理证实的147例乳腺癌病人临床病理及MG、MRI影像资料,按照7∶3比例随机分为训练集(n=102)与测试集(n=45)。在所有病人MG的头尾位(CC)、内外斜位(MLO)及MRI的T2WI、DCE-MRI序列上勾画感兴趣区(ROI)。经最小最大值归一化、Select K Best、LASSO选出与TNBC有较高相关性的最优特征,采用逻辑回归(LR)及支持向量机(SVM)建立基于MG和MRI的多模态影像组学模型,并获取影像组学分数(Rad-score)。通过单及多因素logistic回归得出临床、MG、MRI影像特征独立危险因素构建临床模型。最终将Rad-score联合筛选出的临床-影像危险因素构建多模态影像组学列线图模型。利用受试者工作特征曲线下面积(AUC)评估模型的预测效能,使用校准曲线、决策曲线评估模型的稳定性和临床实用性。结果:多模态影像组学列线图模型预测TNBC的效能最佳,训练集AUC、敏感度、特异度、准确度分别为0.957、90.9%、97.5%、94.1%,测试集分别为0.923、88.9%、91.7%、86.7%。结论:基于MG双体位和MRI双序列的多模态影像组学列线图模型可以在术前较好地、无创预测TNBC。Objective:To investigate the value of multimodal radiomics nomogram model based on mammography(MG)double body position combined with MRI double sequence in preoperative prediction of triple-negative breast cancer(TNBC).Methods:The clinicopathological,MG and MRI imaging data of 147 patients with breast cancer were analyzed,and randomly divided into the training set(n=102)and test set(n=45)according to the ratio of 7∶3.The regions of interest(ROI)were delineated on the cephalic and caudal MG(CC),internal and external oblique(MLO)and T2WI and DCE-MRI sequences of MRI in all patients.After the minimum-maximum normalization,the best features with high correlation with TNBC by Select K Best and LASSO were selected.Logistic regression(LR)and support vector machine(SVM)were used to establish the multimodal radiomics model based on MG and MRI,and the radiomics scores was obtained.The independent risk factors of clinical,MG and MRI image features were obtained by single and multiple logistic regression to construct clinical models.Finally,a multimodal radiomics nomogram model was constructed based on the clinical and imageomics risk factors screened by Rad-score.The area under receiver operating characteristic(AUC)curve was used to evaluate the predictive efficacy of model,and the calibration curve and decision curve were used to evaluate the stability and clinical practicability of model.Results:The efficiency of multimodal radiomics nomogram model in predicting TNBC was the best.The AUC,sensitivity,specificity and accuracy of training set were 0.957,90.9%,97.5%and 94.1%,respectively,and the AUC,sensitivity,specificity and accuracy of test set were 0.923,88.9%,91.7%and 86.7%,respectively.Conclusions:The multimodal radiomics nomogram model based on MG double body position and MRI double sequence can better and noninvasively predict TNBC before surgery.

关 键 词：三阴性乳腺癌 影像组学 多模态 列线图 

分 类 号：R737.9[医药卫生—肿瘤] R445[医药卫生—临床医学]