论文部分内容阅读
目的:探讨基于MR Tn 2加权成像(Tn 2WI)的影像组学标签预测直肠癌KRAS基因突变的潜在价值。n 方法:回顾性研究。纳入山西省肿瘤医院2017年4月—2019年4月行盆腔MR检查并具有KRAS基因检测结果的304例直肠癌患者的临床和影像资料，其中男175例、女129例，中位年龄59.6岁。按7∶3比例将患者随机分为训练组(213例)和验证组(91例)。选取每例患者的高分辨率Tn 2WI进行图像分割及影像组学特征提取，使用单变量统计分析为主的“五步法”进行特征降维，并分别采用多变量logistic回归、决策树(DT)以及支持向量机(SVM)三种分类算法构建影像组学标签，用于预测直肠癌KRAS基因状态。受试者操作特征(ROC)曲线、校正曲线、决策曲线分析(DCA)评估影像组学标签的预测性能及临床效益。n 结果:训练组和验证组患者的基线资料比较以及两组中KRAS突变型与野生型患者的临床特征比较，差异均无统计学意义(n P值均>0.05)。从每位患者的Tn 2WI中提取960个影像组学特征，经特征筛选后得到7个与直肠癌KRAS基因相关的特征(n P值均0.05). A total of 960 features were extracted from the Tn 2WI image of each patient. Seven radiomic features were selected after feature selection, which was significantly associated with KRAS mutations (all n P values<0.05). The area under the ROC curve (AUC) values of the LR, DT, and SVM algorithms were 0.677, 0.604, and 0.722 in the primary cohort, respectively, and 0.626, 0.600, and 0.682 in the validation cohort, respectively. Among the three models, the SVM algorithm showed the best performance for evaluating KRAS mutation, which was validated in the validation cohort with AUCs, sensitivity, specificity, and accuracy of 0.682, 0.713, 0.655, and 0.681, respectively. The DCA curve shows that the three radiomic models have certain clinical benefits, among which the SVM prediction model has the largest net profit.n Conclusions:The radiomic signature based on Tn 2WI exhibits potential for predicting KRAS mutation status in rectal cancer.n