基于双能CT定量参数在胃癌临床分期中的应用

Application of dual-energy CT quantitative parameters in clinical staging of gastric cancer

目的:探讨双能CT定量参数在胃癌临床分期中的应用价值。方法:回顾分析经手术病理证实且术前接受双能CT双期增强检查的200例胃癌患者资料。根据第8版国际抗癌联盟及美国肿瘤联合会胃癌临床TNM分期标准分为Ⅰ+Ⅱ期(95例)与Ⅲ期(105例)两组。在动脉期(AP)和静脉期(VP)图像上分别测量癌灶、胃癌周围脂肪及正常胃周脂肪的双能CT定量参数,包括碘浓度值(IC)、标准化碘浓度值(NIC)、0.6融合图像CT值、静脉期70 keV单能量图像CT值及肿瘤最厚径和最长径。将患者随机分成训练集(150例)和测试集(50例)。采用随机森林(RF)方法来构建模型,通过计算平均下降基尼指数(MDGI)对纳入的变量进行重要性筛选。利用受试者工作特征(ROC)曲线对模型的预测效能进行评估。结果:训练集Ⅲ期组79例、Ⅰ+Ⅱ期组71例,测试集Ⅲ期组26例、Ⅰ+Ⅱ期组24例,RF算法共筛选出5个变量(MDGI>0.13),分别为静脉期胃癌周围脂肪组织IC值(IC_(癌脂)-VP)、最厚径、NIC_(癌脂)-VP、最长径及静脉期70 keV单能量图像胃癌周围脂肪组织CT值(单癌脂-VP)。上述变量构建的RF模型在训练集和测试集中曲线下面积(AUC)分别为0.924(0.882~0.966)、0.917(0.841~0.993)。结论:基于双能CT定量参数构建的RF模型可用于预测胃癌术前临床分期,指导临床决策。

Objective:To assess the application value of quantitative parameters derived from dual-energy computed tomography(DECT)in clinical staging of gastric cancer(GC).Methods:Retrospective analysis was performed in the case data from 200 patients,who underwent DECT scan before operation and were confirmed as GC by surgical pathology.The cases were divided into stageⅠ+Ⅱgroup(n=95)and stageⅢgroup(n=105)by the criteria specified in the 8 th edition of TNM staging system issued by Union for International Cancer Control and American Joint Committee on Cancer(UICC/AJCC).Quantitative parameters were measured at arterial phase(AP)and venous phase(VP)on the images of primary lesions,peritumoral adipose tissue and normal perigastric adipose tissue,including iodine concentration(IC),normalized iodine concentration(NIC),CT values of fused images using weighting factors 0.6,CT values of monoenergetic images at 70 keV,and the longest and thickest tumor diameter.All patients were randomized to training set(n=150)and test set(n=50).Random forest(RF)algorithm was used to construct the model,and the included variables were screened for importance by calculating the mean decrease in Gini index(MDGI).The receiver operating characteristic(ROC)curve was applied to evaluate the predictive efficiency of the model.Results:The training set included 79 cases in stageⅢgroup and 71 cases in stageⅠ+Ⅱgroup,and the test set was constituted by 26 cases in stageⅢgroup and 24 cases in stageⅠ+Ⅱgroup.Five variables(MDGI>0.13)were screened out via RF algorithm,including the IC-VP of peritumoral adipose tissue,the thickest diameter,NIC-VP of peritumoral adipose tissue,the longest diameter,and VP 70 keV monoenergetic image values of peritumoral adipose tissue.The area under curve(AUC)of the RF model constructed with the above variables in the training set and the test set was 0.924(0.882-0.966)and 0.917(0.841-0.993),respectively.Conclusion:RF model based on the quantitative parameters generated from DECT can be used to predict preoperative clinical staging of gastric cancer and guide clinical decision-making.