【摘要】 目的探索多参数MRI影像组学信号模型预测直肠癌(rectal cancer,RC) KRAS基因突变的价值。材料与方法回顾性分析深圳市人民医院2019年4月至2020年12月104例经病理证实且行术前MRI检查的直肠癌患者的临床病理资料和提取RC的多参数MRI影像组学特征。采用t检验、χ2检验或Mann-Whitney U检验分析临床病理特征和影像组学特征与KRAS基因突变的相关性,将有统计学意义的特征纳入LASSO回归模型进行特征选择和建立影像组学信号。采用受试者工作特征(receiver operating characteristic,ROC)的曲线下面积(area under the curve,AUC)评价影像组学信号对KRAS基因突变的预测效能。结果临床病理资料在有无KRAS基因突变间差别无统计学意义。321个影像组学特征中,单因素分析表明16个影像组学特征与KRAS基因突变有相关性。LASSO回归筛选出7个影像组学特征构建影像组学信号,在验证集和预测集中预测KRAS基因突变的AUC值分别为0.81 (0.70～0.92)和0.77 (0.63～0.91,P=0.60),其中ADC特征中一阶偏度的压缩系数最大为3.36。结论 MRI影像组学特征可以作为预测KRAS基因突变的生物学标记,其中ADC特征中偏度的预测效能最好。更多还原

【Abstract】 Objective: To explore the value of multiparametric MRI imaging omics signal model to predict KRAS gene mutation in rectal cancer(RC). Methods and Materials: The clinicopathological data and the multi-parameter MRI imaging features of 104 patients with histopathological proven RC and preoperative MRI were retrospective recruited from Apr. 2019 to Dec. 2020. The association of clinicopathological characteristics and radiomics features with KRAS gene were evaluated using t test, χ2 test or Mann-Whitney U test.Least absolute shrinkage and selection operator(LASSO) regression was harnessed for radiomics features selection and radiomics signature building. Prediction performance of radiomics signature for KRAS gene mutation was assessed by using area under the curve(AUC) of receiver operating characteristic(ROC). Results: The associations of clinicopathological characteristics and KRAS gene mutation were not statistical significant. Univaraite analysis revealed that 16 of the 321 radiomics features were related to KRAS mutation. LASSO regression selected 7 features for radiomics signature building. The radiomics signature yielded AUC of 0.81(95% CI:0.70—0.92) and 0.77(95% CI: 0.63—0.91, P=0.60) for predicting KRAS mutation in training and validation sets, among them, the maximum λ coefficiences of the first-order skewness in the ADC feature is 3.36. Conclusions: MRI radiomics signature could be used as surrogate biomarker for predicting KRAS mutation in RC, among them, the first-order skewnes of ADC features has the best predictive performance.更多还原