多模态影像组学列线图在预测乳腺癌患者HER-2表达水平中的价值  

Application of Multimodal Radiomics Nomogram in Predicting HER-2 Expression in Breast Cancer Patients

作　　者：杨静茹 赵楠楠 张舒妮 朱芸 张澳祺 李阳 顾一泓 谢宗玉 YANGJingru;ZHAO Nannan;ZHANG Shuni(Graduate School of Bengbu Medical University,Bengbu,Anhui Province 233030,P.R.China)

机构地区：[1]蚌埠医科大学第一附属医院放射科,233004 [2]蚌埠医科大学研究生院,233030

出　　处：《临床放射学杂志》2024年第9期1471-1477,共7页Journal of Clinical Radiology

基　　金：安徽省重点研究与开发计划项目(编号:2022e07020033);安徽省教育厅自然科学重点项目(编号:2022AH051473);蚌埠医学院临床研究专硕(编号:2022byflc008);蚌埠医学院研究生科研创新项目(编号:Byycx23069)。

摘　　要：目的探讨基于多模态影像组学联合临床特征构建的列线图预测乳腺癌患者人表皮生长因子受体2(HER-2)的表达水平。方法回顾性分析蚌埠医科大学第一附属医院术前经磁共振(MRI)和钼靶(MG)检查且经病理证实的201例乳腺癌患者,以7∶3的比例随机分配为训练集(n=140)和测试集(n=61)。选择MG头尾(CC)位、内外斜(MLO)位及T_(2)WI脂肪抑制(FS-T_(2)WI)序列、DCE-MRI第2期勾画病灶最大层面。通过f-calssif函数、最小绝对收缩和选择算子(LASSO)回归筛选最优特征。通过支持向量机(SVM)获取钼靶(MG)、动态对比增强MRI(DCE-MRI)、多模态影像组学评分(Rad-score),分别构建MG模型、MRI模型、多模态模型。通过单-多因素Logistic回归筛选临床独立预测因子构建临床模型,选择多模态模型Rad-score联合临床独立预测因素建立列线图模型。结果列线图模型其训练集AUC、敏感度、特异度及准确度分别为0.902、91.8%、85.7%及85.0%;测试集分别为0.886、81.8%、84.6%及80.3%。结论列线图有可能作为一种准确、无创方法用于预测术前乳腺癌患者HER-2表达水平,为临床诊疗和决策提供重要指导。Objective To investigate the prediction of human epidermal growth factor receptor 2(HER-2)expression level in breast cancer patients based on the nomogram constructed by multimodal radiomics combined with clinical features.Methods 201 patients with pathologically confirmed breast cancer who underwent magnetic resonance imaging(MRI)and mammogram(MG)examination before surgery in the First Affiliated Hospital of Bengbu Medical University were analyzed retrospectively,and randomly assigned to the training set(n=140)and test set(n=61)at a ratio of 7∶3.MG craniocaudal(CC)position,Mediolateral oblique(MLO)position,T_(2)WI fat inhibition(FS-T_(2)WI)sequence and DCE-MRI phase 2 were selected to delineate the maximum lesion level.The optimal features were selected by f-calssif function,least absolute shrinkage and selection operator(LASSO)regression,and mammogram(MG),dynamic contrast-enhanced MRI(DCE-MRI)and multimodal radiomics score(Rad-score)were obtained by support vector machine(SVM).MG model,MRI model and multi-modal model were constructed respectively.The independent clinical predictors were screened by single-multiple logistic regression to build the clinical model,and the multi-modal model Rad-score combined with independent clinical predictors was selected to build the nomogram model.Results The AUC,sensitivity,specificity and accuracy of the training set of nomogram model were 0.902,91.8%,85.7%and 85.0%,respectively.The test sets were 0.886,81.8%,84.6%and 80.3%,respectively.Conclusion The nomogram may be used as an accurate and non-invasive method to predict the expression level of HER-2 in preoperative breast cancer patients,and provide important guidance for clinical diagnosis and treatment and decision-making.

关 键 词：乳腺癌 HER-2表达 多模态 影像组学 列线图 

分 类 号：R737.9[医药卫生—肿瘤] R445.2[医药卫生—临床医学] R730.44