基于MRI影像组学模型与临床模型预测乳腺癌新辅助化疗病理完全缓解效能分析  

MRI-based radiomics and clinical-based model for predicting the complete pathological response to neoadjuvant chemotherapy in breast cancer

作　　者：程凤燕[1] 杨志企[1] 廖玉婷 杨佳达 陈湘光[1] 范伟雄[1] 陈小凤 CHENG Fengyan;YANG Zhiqi;LIAO Yuting;YANG Jiada;CHEN Xiangguang;FAN Weixiong;CHEN Xiaofeng(Department of Radiology,Meizhou People’s Hospital,Meizhou 514031,China;GE Healthcare,Meizhou People’s Hospital,Meizhou 514031,China)

机构地区：[1]梅州市人民医院放射科,梅州514031 [2]GE医疗,梅州514031

出　　处：《国际医学放射学杂志》2021年第4期398-402,414,共6页International Journal of Medical Radiology

基　　金：梅州市社会发展科技计划项目(2020B105);广东省医学科研基金项目(B2021280)。

摘　　要：目的比较基于动态增强磁共振成像(DCE-MRI)、表观扩散系数(ADC)图的影像组学模型以及基于临床特征模型预测乳腺癌新辅助化疗(NAC)病理完全缓解(PCR)的效能。方法回顾性收集91例行乳腺癌NAC并有疗效病理评估结果的女性病人,平均年龄(48.45±9.49)岁。将91例病人按照7∶3比例分为训练组(63例)和验证组(28例)。2组均进行NAC疗效病理评估,训练组中PCR者16例、病理部分缓解(PPR)者47例,验证组中PCR者7例、PPR者21例。所有病人均在NAC前行DCE-MRI和扩散加权成像(DWI)检查。采用单因素Logistic回归对病人年龄、雌激素受体(ER)、孕激素受体(PR)、人表皮生长因子受体2(HER-2)、肿瘤增殖细胞核抗原(Ki-67)表达状态进行分析,然后基于有统计学意义的临床特征建立临床模型。提取并筛选影像组学特征,采用多元Logistic回归构建DCE模型和ADC模型,计算相应模型影像组学评分(DCE-Radscore和ADC-Radscore)。采用t检验、卡方检验或Fisher确切概率检验比较训练组和验证组中PCR者和PPR者间的临床特征和影像组学评分。采用受试者操作特征(ROC)曲线评估模型的预测效能,并计算其敏感度、特异度和曲线下面积(AUC)。采用决策曲线评估模型的临床应用价值。结果训练组中,PCR者的ER、PR阴性率、DCE-Radscore均高于PPR者(均P<0.05),并在验证组中得到验证(P<0.05)。训练组中PCR者的ADC-Radscore高于PPR者(P<0.05),但未得到验证组验证(P>0.05)。训练组中,临床模型预测PCR的AUC值(0.823)及敏感度(0.875)最高,其次是ADC模型,DCE模型最低(AUC为0.750,敏感度为0.688)。验证组中,临床模型预测PCR的AUC值最高(0.793)而敏感度(0.571)最低,ADC模型AUC(0.639)最低但敏感度(1.000)最高。决策曲线分析显示应用临床模型具有最大的净获益,其次是ADC模型,DCE模型最低。结论ADC模型、DCE模型和临床模型均能预测PCR,其中临床模型预测效能和净获益最高。Objective To explore the efficiency of radiomics model based on dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)and ADC map,and clinical model based on clinical features in predicting the pathological complete response(PCR)to neoadjuvant chemotherapy(NAC)in breast cancer.Methods All of 91 breast cancer women who had received NAC and had pathological response results of NAC were retrospective analysis.The mean age was 48.45±9.49 years.A training cohort consisted of 63 patients and a testing cohort consisted of 28 patients at a rate of 7:3.According to the pathological response results of NAC,16 patients had PCR and 47 patients had pathologic partial response(PPR)in the training cohort,and 7 patients had PCR and 21 patients had PPR in the testing cohort.All the patients underwent DCE-MRI and diffusion weighted imaging before NAC.Univariate logistic regression analysis was run for the following clinical features:age,ER,PR,HER-2,and Ki-67 status.Then these features showing statistical significance were fed to develop the clinical model.A list of radiomics features were extracted and selected,multivariable logistic regression was used to develop the DCE model and ADC model.and corresponding Radscores inculidng DCE-Radscore and ADC-Radscore were calculated.The t test,χ2,or Fisher test was used to compared the clinical features and Radscores between PCR and PPR groups in the training and testing cohort.The predictive ability of the models were validated by using receiver operating characteristic(ROC)analysis,and the area under the ROC curve(AUC),accuracy,sensitivity,and specificity were calculated.The decision curve was conducted to determine the clinical usefulness of the models.Results PCR patients had higher positive rate of ER and PR and DCE-Radscore than PPR patients in the training cohort(P<0.05),which was then confirmed in the testing cohort(P<0.05).PCR patients had higher ADC-Radscore than PPR patients in the training cohort(P<0.05),but this was not confirmed in the testing cohort(P>0.05).In the t

关 键 词：乳腺癌 磁共振成像 影像组学 新辅助化疗 诊断效能 

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