CT影像组学联合传统影像特征对胃肠道间质瘤Ki-67指数表达的预测价值

Predictive value of contrast-enhanced CT radiomics combined with traditional imaging features for Ki-67 index expression in gastrointestinal stromal tumors

目的:探讨基于增强CT的影像组学联合传统影像特征对无创性预测胃肠道间质瘤(GIST)Ki-67增殖指数(Ki-67 PI)表达的价值。方法:回顾性搜集我院2010年9月至2020年9月经手术病理确诊的原发性GIST患者的病例资料。诊断医师采用盲法独立分析增强CT图像,提取传统的影像征象。利用ITK-SNAP软件在增强CT图像上勾画病灶感兴趣区,利用AK软件提取纹理特征,将患者随机分为训练集与验证集。采用逻辑回归筛选特征参数并构建影像组学模型。再分别建立传统影像特征模型、组学模型及联合两者的组合模型。结果:传统影像特征模型的诊断效能尚可,其ROC曲线的曲线下面积(AUC)在训练集和验证集中分别为0.720(95%CI:0.651~0.788)及0.665(95%CI:0.547~0.784)。影像组学模型的诊断效能良好,其AUC在训练集中具有最优值,为0.802(95%CI:0.744~0.860),其AUC在验证集中为0.730(95%CI:0.623~0.836)。此外,联合影像组学和传统影像特征组成的多参数组合模型在训练集中效能良好,AUC值为0.823(95%CI:0.768~0.878),其在验证集中具有最优的诊断效能,AUC值为0.731(95%CI:0.626~0.836)。结论:基于增强CT的影像组学联合传统影像特征建立的组合模型具有无创预测GIST患者Ki-67 PI表达状态的价值。

Objective:To investigate the value of contrast-enhanced computed tomography(CECT)radiomics combined with traditional imaging features in noninvasive prediction of Ki-67 proliferation index(Ki-67 PI)expression in gastrointestinal stromal tumors(GISTs).Methods:The data of patients with primary GISTs diagnosed by surgical pathology in our hospital from September 2010 to September 2020 were retrospectively collected.The diagnostic radiologists were blinded to independently analyze the CECT images to record the traditional imaging features.Patients were randomly divided into training and validation groups using ITK-SNAP software to outline the region of interest(ROI)of the tumor on CECT images,and using AK software to extract texture features.Logistic regression was used to screen the feature parameters and construct the radiomics model.Then,the traditional imaging features model,the radiomics model and the combination model were established respectively.Results:The results showed that the diagnostic performance of the traditional imaging features model was average,with the area under curve(AUCs)values of 0.720[95%confidence interval(CI):0.651 to 0.788)and 0.665(95%CI:0.547 to 0.784)]in the training and validation groups,respectively.The diagnostic efficacy of the radiomics model was good,with an optimal AUC value of 0.802(95%CI:0.744 to 0.860)in the training group and an AUC value of 0.730(95%CI:0.623 to 0.836)in the validation group.In addition,the multiparametric combination model consisting of radiomics and traditional imaging features had good efficacy in the training group with an AUC value of 0.823(95%CI:0.768 to 0.878),and had the best diagnostic performance in the validation group with an AUC value of 0.731(95%CI:0.626 to 0.836).Conclusion:The combination model based on CECT radiomics combined with traditional imaging features has the value of noninvasively predicting Ki-67 PI expression in GISTs.