基于CT影像组学预测结直肠癌KRAS基因突变

CT-Based Radiomic Model for Predicting KRAS Gene Mutation Status in Colorectal Cancer

目的探讨基于CT影像组学模型术前预测结直肠癌KRAS基因状态的价值。资料与方法回顾性收集2019年7月—2022年1月四川省肿瘤医院经病理证实的原发性结直肠癌177例,分为训练组123例(62例KRAS突变,61例KRAS野生型)和验证组54例(31例KRAS突变,23例KRAS野生型)。以每例结直肠癌的原发灶作为感兴趣区,共提取1352个放射学特征。使用最小冗余最大相关、最小绝对收缩和选择算子、向后逐步Logistic回归3个步骤在训练组中筛选关键特征,并建立KRAS预测的影像组学模型。通过受试者工作特征曲线下面积评价预测性能。绘制校准曲线以评估预测概率和实际概率之间的一致性。使用Hosmer-Lemeshow检验评估影像组学模型的拟合优度。通过决策曲线分析评估放射模型的应用价值。结果选择7个特征建立影像组学模型,该模型在突变KRAS组和野生型KRAS组之间的区分性能显示,训练组的曲线下面积为0.755(95%CI 0.669~0.828),验证组的曲线下面积为0.724(95%CI 0.585~0.837)。校准曲线显示,预测概率和实际概率之间具有良好的一致性,且Hosmer-Lemeshow测试结果显示在训练组和验证组中差异均无统计学意义(χ^(2)=8.427,P=0.310;χ^(2)=6.054,P=0.630)。临床决策曲线显示,当风险阈值为30%~98%时,采用影像组学方法预测KRAS的临床获益较高。结论基于CT的影像组学模型术前预测结直肠癌KRAS突变具有一定的诊断价值。

Purpose To develop a CT-based radiomic model for preoperative prediction of KRAS mutation status in colorectal cancer.Materials and Methods 177 patients with confirmed primary colorectal cancer were retrospectively collected in Sichuan Cancer Hospital from July 2019 to January 2022,and further divided into primary cohort(123 cases,62 with mutant KRAS and 61 with wild-type KRAS)and validation cohort(54 cases,31 with mutant KRAS and 23 with wild-type KRAS).A total of 1352 radiomic features were extracted from the regions of interest of colorectal cancer based on CT images.Three steps including the minimum redundancy-maximum relevance,least absolute shrinkage and selection operator regression,backward stepwise multivariable Logistic regression were applied to select key features,and developed the radiomic model for KRAS prediction in the primary cohort.The predictive performance was evaluated by area under the receiver operating characteristic curve.The calibration curve was plotted to assess the consistency between predicted and actual probability,and the Hosmer-Lemeshow test was performed to estimate the goodness-of-fit of the radiomic model.The clinical application of the radiomic model was accessed through the decision curve analysis.Results The radiomic model included 7 selected features and showed a good discrimination performance between the mutant KRAS and wild-type KRAS groups,with an area under the curve of 0.755(95%CI 0.669-0.828)in the primary cohort and 0.724(95%CI 0.585-0.837)in the validation cohort.The calibration curve showed good consistency between prediction and observation,and the Hosmer-Lemeshow test showed no significant difference in both the primary and validation cohorts(χ^(2)=8.427,P=0.310;χ^(2)=6.054,P=0.630).The decision curve analysis showed the radiomic model has a good clinical utility when the risk threshold was 30%-98%.Conclusion CT-based radiomic model has predictive value to preoperatively predict the KRAS mutation status in colorectal cancer to some extent.