基于CT影像组学数据构建的综合模型预测乳腺癌新辅助化疗疗效价值的研究  

VALUE OF A CT-BASED RADIOMICS NOMOGRAM IN PREDICTING THE RESPONSE OF BREAST CANCER TO NEOADJUVANT CHEMOTHERAPY

作　　者：段艳莉 杨光杰[1] 王洋洋[1] 毛艳[2] 苗文杰 颜蕾 左盼莉 金萌 麻杰 李晓旭 王振光[1] DUAN Yanli;YANG Guangjie;WANG Yangyang;MAO Yan;MIAO Wenjie;YAN Lei;ZUO Panli;JIN Meng;MA Jie;LI Xiaoxu;WANG Zhenguang(PET-CT Center, The Affiliated Hospital of Qingdao University, Qingdao 266061, China)

机构地区：[1]青岛大学附属医院PET-CT中心,山东青岛266061 [2]青岛大学附属医院乳腺病诊疗中心 [3]慧影医疗科技(北京)有限公司

出　　处：《精准医学杂志》2021年第6期477-482,487,共7页Journal of Precision Medicine

基　　金：山东省中医药科技发展计划项目(2017-193);青岛市市南区科技计划项目(2020-2-004-YY)。

摘　　要：目的建立并验证基于CT影像组学的综合模型预测乳腺癌新辅助化疗(NAC)疗效的价值。方法回顾性分析2017年1—12月109例于我院行NAC并具有MP病理分级结果的乳腺癌患者,根据CT检查时间分为训练集77例(检查时间:2017年1—9月)和验证集32例(检查时间:2017年10—12月)。采用ITK-SNAP软件勾画肿瘤三维感兴趣区(ROI)并提取影像组学特征,采用mRMR算法和LASSO回归进行最佳影像组学特征选择并构建影像组学模型,并计算每位患者的影像组学得分。再通过多元Logistic回归分析分别构建临床特征模型与联合影像组学得分和临床特征的综合模型,并绘制列线图。采用校正曲线评估模型的拟合度。通过ROC曲线评价各模型预测乳腺癌NAC疗效的效能,并通过Delong检验比较临床特征模型与综合模型预测效能是否具有统计学差异。最后再通过决策曲线评价综合模型预测乳腺癌NAC疗效的净获益。结果共提取1409个影像组学特征,最终得到12个最佳的影像组学特征。影像组学特征对预测乳腺癌NAC疗效具有较好的效能,其训练集和验证集的ROC曲线下面积(AUC)分别为0.83和0.76,联合临床特征和影像组学得分构建的综合模型在训练集和验证集中预测NAC疗效的AUC分别为0.88和0.78。Delong检验结果表明在训练集和验证集中,临床特征模型与综合模型预测效能具有统计学差异(训练集Z=2.922,P<0.05;验证集Z=2.318,P<0.05)。决策曲线表明基于CT影像组学的综合模型预测乳腺癌NAC疗效较临床模型具有较高的净获益。结论基于CT影像组学的综合模型对乳腺癌NAC疗效具有较高的预测价值,有助于指导乳腺癌患者个体化诊疗。Objective To establish a CT-based radiomics nomogram,and to investigate its value in predicting the response of breast cancer to neoadjuvant chemotherapy(NAC).Methods A retrospective analysis was performed for 109 patients who received NAC and had MP pathological classification results in our hospital from January to December 2017.According to the time of CT examination,the patients were divided into training set with 77 patients(time of examination:January to September 2017)and validation set with 32 patients(time of examination:October to December 2017).ITK-SNAP software was used to delineate the three-dimensional region of interest(ROI)of the tumor and extract radiomics features;the mRMR algorithm and LASSO regression were used to select the optimal radiomics features and establish a radiomics model;the radiomics score was calculated for each patient.A logistic regression analysis was used to establish a clinical feature model and a radiomics nomogram which incorporated clinical features and radiomics score,and a calibration curve was used to evaluate the degree of fitting of the models.The receiver operating characteristic(ROC)curve was used to evaluate the performance of each model in predicting the response of breast cancer to NAC,and the Delong test was used to compare the difference in the performance of the two models.Finally,the decision curve analysis(DCA)was performed to evaluate the net benefits of the models in predicting the response of breast cancer to NAC.Results A total of 1409 radiomics features were extracted,and 12 optimal radiomics features were obtained.Radiomics features had good performance in predicting the response of breast cancer to NAC,with an area under the ROC curve(AUC)of 0.83 in the training set and 0.76 in the validation set.The radiomics nomogram combining clinical features with radiomics score had an AUC of 0.88 in the training set and 0.78 in the validation set in predicting the response of breast cancer to NAC.The Delong test showed a significant difference in predictive per

关 键 词：图像处理 计算机辅助 体层摄影术 X线计算机 计算机模拟 影像组学 乳腺肿瘤 肿瘤辅助疗法 化学疗法 辅助 疗效预测 

分 类 号：R737.9[医药卫生—肿瘤] R319[医药卫生—临床医学]