目的 探究基于机器学习组合模型的影像组学在预测肿块型乳腺癌新辅助化疗(neoadjuvant chemotherapy,NAC)疗效中的价值.材料与方法 回顾性分析2018年1月到2021年10月中国人民解放军总医院第五医学中心的97例接受NAC治疗且经组织病理学证实的肿块型乳腺癌患者的临床和影像资料.基于实体瘤疗效评定标准(Response Evaluation Criteria in Solid Tumors,RECIST),将患者分为有效组和无效组,基于治疗前的动态对比增强MRI(dynamic contrast-enhanced MRI,DCE-MRI)减影第一期图像上提取的影像组学特征,引入高通或低通小波滤波器和不同参数的拉普拉斯-高斯滤波器对原始MR图像进行预处理.采用基于单变量分析和多变量分析的特征选择方法进行特征筛选,单变量分析包括F检验、卡方检验和互信息;多变量分析采用最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO);采用支持向量机(support vector machine,SVM)、随机森林(random forest,RM)、logistic回归分析(logistic regression,LR)三种机器学习分类器进行建模,通过交叉组合,共有12种特征筛选器和分类器的组合方案,训练时采用十次重复的五折交叉验证.使用曲线下面积(area under the curve,AUC)、敏感度、特异度、准确率,阳性预测值和阴性预测值评估模型的预测能力.结果 在所有交叉组合的方案中,最佳的特征筛选器是单变量分析中的F检验方法,最佳的分类器是SVM模型,该组合共筛选出191个影像组学特征,AUC为0.83[95%置信区间(confidence interval,CI):0.80～0.86],准确率为 77%(95%CI:74%～80%),特异度为 81%(95%CI:78%～84%),敏感度为71%(95%CI:65%～77%),阳性预测值为67%(95%CI:62%～72%),阴性预测值为85%(95%CI:83%～87%).结论 基于F检验方法和SVM的机器学习组合模型证实了影像组学在预测肿块型乳腺癌NAC疗效中有一定的价值.
Radiomics based on combined machine learning models for prediction of the response to neoadjuvant chemotherapy in mass enhancement breast cancer using magnetic resonance imaging
Objective:To investigate the value of radiomics based on combined machine learning models in predicting the response to neoadjuvant chemotherapy(NAC)in mass enhancement breast cancer.Materials and Methods:The clinical and imaging data of ninety-seven patients with mass enhancement breast cancer confirmed by histopathology and underwent NAC from January 2018 to October 2021 in the Fifth Medical Center of Chinese PLA General Hospital were retrospectively analyzed in our study.Based on the results of Response Evaluation Criteria in Solid Tumors(RECIST),the patients were classified into effective group and ineffective group.Based on the radiomics features extracted on the first dynamic contrast-enhanced MRI(DCE-MRI)subtraction images before treatment,a high-pass or low-pass wavelet filter and a Laplace-Gaussian filter with different parameters were also introduced to preprocess the original MR images.For feature screening,feature selection methods based on univariate analysis and multivariate analysis were used.The univariate analysis included F-test,chi-square test and mutual information.The multivariate analysis used the least absolute shrinkage and selection operator(LASSO).Support vector machine(SVM),random forest(RM),and logistic regression(LR)were used for modeling,and finally a total of twelve combinations of feature filters and classifiers were combined by crossover.Ten repetitions of five-fold cross-validation were used for training.Finally,area under the curve(AUC),sensitivity,specificity,accuracy,positive prediction value and negative prediction value were used to evaluate the prediction performance.Results:Among all cross-combined schemes,the feature screening method that achieved the best classification performance was the F-test method in univariate analysis,and the best classifier was the SVM.The combination screened a total of 191 imaging features with an overall mean AUC of 0.83[95%confidence interval(CI):0.80-0.86]in predicting NAC response,and the accuracy of the model was 77%(95%CI:74%-80%).Specificity was 81%(95%CI:78%-84%),sensitivity was 71%(95%CI:65%-77%),positive predictive value was 67%(95%CI:62%-72%),and negative predictive value was 85%(95%CI:83%-87%).Conclusions:A combined machine learning model based on F-test and SVM validated good performance of radiomics in predicting the response to NAC for mass enhancement breast cancer patients.
breast cancerneoadjuvant chemotherapyradiomicsmachine learningmagnetic resonance imaging
国家科技期刊平台
NSTL
万方数据
