MRI影像组学预测结直肠癌患者KRAS基因突变

MRI radiomics for predicting KRASgene mutation in colorectal cancer patients

目的构建MR T2WI影像组学模型,评价其预测结直肠癌患者Kirsten大鼠肉瘤(KRAS)病毒癌基因亚型的价值。方法将99例经病理证实的结直肠癌患者分为训练组(n=68)及验证组(n=31),根据KRAS基因检测结果进一步将其分为突变亚组及野生亚组,训练组2亚组分别含36、32例,验证组2亚组分别含16、15例,比较亚组间实验室检查结果及肿瘤大小的差异;提取并筛选训练组MR T2WI影像组学特征,构建影像组学模型、临床模型及影像组学-临床联合模型。绘制受试者工作特征(ROC)曲线,计算曲线下面积(AUC),评价各模型预测结直肠癌患者KRAS基因亚型的效能;以DeLong检验比较各模型间效能差异。通过校正曲线分析3种模型的校正性能,以Hosmer-Lemeshow检验评价校准曲线的校准度;以决策曲线分析(DCA)评价3种模型临床应用价值。结果训练组和验证组内亚组间实验室检查结果及肿瘤大小差异均无统计学意义(P均>0.05)。共提取3个组学特征用于构建预测模型。影像组学模型与临床模型、影像组学-临床联合模型预测2组KRAS基因亚型的AUC差异均无统计学意义(P均>0.05);影像组学-临床联合模型预测训练组KRAS基因亚型的AUC显著高于临床模型(P<0.05),但在验证组差异均无统计学意义(P>0.05)。校准曲线及Hosmer-Lemeshow检验显示3种模型预测值和观察值的一致性良好(P均>0.05)。影像组学模型和影像组学-临床联合模型在2组中的DCA曲线净收益值均高于临床模型。结论MR T2WI影像组学纹理特征预测结直肠癌患者KRAS基因突变亚型具有一定潜力。

Objective To construct MR T2WI radiomics model and investigate its value for predicting Kirsten rat sarcoma(KRAS) virus oncogene subtype in patients with colorectal cancer. Methods Totally 99 patients with colorectal cancer confirmed by pathology were divided into training group(n=68) and verification group(n=31), and then further divided into mutation subgroup and wild subgroup according to the KRAS gene test results. There were 36 and 32 cases in mutation subgroup and wild subgroup of training group, and 16 and 15 cases in those of validation group, respectively. The laboratory examination results and tumor sizes were compared among subgroups. MR T2WI radiomics features were extracted and screened from training group, and then the radiomics model, clinical model and combining radiomics-clinical model were conducted. The receiver operating characteristic(ROC) curve was drawn, and the area under the curve(AUC) was calculated to evaluate the effectiveness of each model for predicting the KRAS gene subtype of colorectal cancer. DeLong test was used to compare the effectiveness among 3 models. The calibration performances of the models were analyzed with calibration curve, and the calibration degree of the calibration curve was evaluated with Hosmer-Lemeshow test. The clinical application value of the models was evaluated with decision curve analysis(DCA). Results Within training group and validation group, no significant difference of laboratory examination results nor tumor sizes was found between 2 subgroups(all P>0.05). Three radiomics features were extracted for constructing the prediction model. No significant difference of AUC between radiomic model and clinical model nor between radiomics model and combining radiomics-clinical model for predicting the KRAS gene subtypes of colorectal cancer was found in both groups(all P>0.05). In training group, AUC of combining radiomics-clinical model for predicting KRAS gene subtype was significantly higher than that of clinical model(P<0.05), but there was no significant difference of AUC in validation group(P>0.05). The calibration curves and the Hosmer-Lemeshow test of 3 models showed good consistency between the predicted values and the observed values(all P>0.05). The net income value of DCA curve of radiomics model and combining radiomics-clinical model were higher than that of clinical model in both groups. Conclusion MR T2WI radiomics texture features had certain potential for predicting KRAS gene mutation subtypes in patients with colorectal cancer.