影像组学模型预测胃肠道间质瘤危险度分级的价值

The value of radiomics model in predicting the risk grade of gastrointestinal stromal tumors

目的:探讨影像组学模型预测原发性胃肠道间质瘤危险度分级的价值。方法:回顾性分析2017年1月—2020年12月间苏州大学附属第二医院收治的经手术病理证实为原发性胃肠道间质瘤的176例患者,分为低度恶性潜能组83例,高度恶性潜能组93例。选取各期CT图像中肿瘤最大层面图像,使用Mazda软件进行特征提取,采用LASSO算法构建模型。使用C统计量和Brier评分评价模型区分度及校准度,利用加强Bootsrap法对模型进行内部验证。计算模型的特异度、灵敏度及准确率。结果:最终10个特征被纳入模型,模型C统计量为0.943,95%置信区间为0.911~0.976,校准后的C统计量为0.928;Brier评分为0.093,95%置信区间为0.065~0.121,校准后的Brier评分为0.112;模型特异度为0.892,灵敏度为0.860,准确率为0.875。结论:基于CT图像构建的影像组学模型对间质瘤危险度分级具有较高的预测价值,可为临床医师制定治疗方案提供帮助。

Objective To explore the value of radiomics model in predicting the risk grade of primary gastrointestinal stromal tumors(GIST).Methods Retrospective analysis of 176 patients with primary gastrointestinal stromal tumors confirmed by surgery and pathology admitted to the Second Affiliated Hospital of Suzhou University from January 2017 to December 2020.83 cases were divided into low malignant potential group and 93 cases were in high malignant potential group.Select the largest slice image of GIST in each phase of CT image,Mazda software was used for feature extraction.Then LASSO algorithm was used to obtain the radiomics model.C-index and Brier score were used to evaluate model differentiation and calibration.The enhanced Bootsrap method was used to verify the model internally.And the specificity,sensitivity and accuracy of the model were calculated.Results Finally,10 features were included in the model,and the C-index of the model was 0.943,the 95%CI was 0.911-0.976,the Brier score was 0.093,and the 95%CI was 0.065-0.121,and the calibrated C-index was 0.928,and the calibrated Brier score was 0.112.The specificity,sensitivity and accuracy of the model were 0.892,0.860 and 0.875.Conclusion The radiomics model constructed based on CT images has a high predictive value for GIST risk classification,it can be used to provide help for clinicians to make treatment plans.