CT影像组学预测合并腹膜转移晚期胃癌新辅助化疗初期疗效的研究  

Value of CT radiomics in predicting the initial efficacy of neoadjuvant chemotherapy for advanced gastric cancer with peritoneal metastasis

作　　者：陈智成 朱可心 赵金莉 刘屹[1] CHEN Zhicheng;ZHU Kexin;ZHAO Jinli;LIU Yi(Department of Radiology,the First Hospital of China Medical University,Shenyang 110001,China)

机构地区：[1]中国医科大学附属第一医院放射科,辽宁沈阳110001

出　　处：《实用放射学杂志》2022年第3期420-423,436,共5页Journal of Practical Radiology

摘　　要：目的探讨CT影像组学模型在预测合并腹膜转移晚期胃癌(AGC)新辅助化疗初期疗效的临床价值.方法回顾性分析93例经病理证实为腹膜转移的AGC患者的影像资料,其中化疗疗效明显48例[部分缓解(PR)],化疗疗效不明显45例[疾病进展(PD)+疾病稳定(SD)].使用ITK-SNAP软件在门静脉期CT图像上手动勾画感兴趣区(ROI)并导入A.K.软件进行特征提取.将数据按7︰3随机分为训练集和验证集,通过R软件中的“corrplot”、“Boruta”包进行特征筛选.利用筛选得到的特征分别构建4种影像组学模型,每个影像组学模型均采用10折交叉验证并绘制各模型的受试者工作特征(ROC)曲线以评价模型预测效能.结果筛选得到18个影像组学特征,基于这些特征构建的模型的曲线下面积(AUC)值分别为0.85[线性判别分析(LDA)]、0.73[随机森林(RF)、0.72[支持向量机(SVM)]和0.66[K邻近(KNN)].结论CT影像组学模型在预测合并腹膜转移AGC患者新辅助化疗初期疗效方面潜力巨大,其中LDA影像组学模型为最佳预测模型,对帮助临床决策及实现个体化治疗具有重要参考价值.Objective To discuss the clinical value of CT radiomics models in predicting the initial efficacy of neoadjuvant chemotherapy for advanced gastric cancer(AGC)with peritoneal metastasis.Methods The imaging data of 93 AGC patients with peritoneal metastasis confirmed by pathology were analyzed retrospectively,including 48 cases[partial response(PR)]with obvious effect of chemotherapy and 45 cases[progressive disease(PD)+stable disease(SD)]with no obvious effect of chemotherapy.ITK-SNAP software was used to manually delineate the region of interest(ROI)on the portal phase CT images,and then imported to A.K.software for feature extraction.All the data were randomly divided into training group and validation group according to the ratio of 7︰3,and feature screening was carried out through“corrplot”and“Boruta”packages in R software.Four radiomics models were constructed by using the selected features,and each radiomics model was cross-validated by ten folds.The receiver operating characteristic(ROC)curve of each model was drawn to evaluate the predictive performance.Results 18 radiomics features were selected,and the area under the curve(AUC)values of the radiomics models were 0.85[linear discriminant analysis(LDA)],0.73[random forest(RF)],0.72[support vector machine(SVM)]and 0.66[K nearest neighbor(KNN)],respectively.Conclusion CT radiomics models have great potential in predicting the initial efficacy of neoadjuvant chemotherapy for AGC patients with peritoneal metastasis,and the LDA radiomics model is the best prediction model,which provides an important reference for helping clinical decision-making and realizing individualized treatment.

关 键 词：计算机体层成像 胃癌 新辅助化疗 影像组学 

分 类 号：R814.42[医药卫生—影像医学与核医学] R735.2[医药卫生—放射医学] R818.051[医药卫生—临床医学]