基于剂量组学的食管癌放化疗后放射性肺炎的预测模型  

Prediction model of radiation pneumonitis after chemoradiotherapy for esophageal cancer based on dosiomics

作　　者：白雪[1,2,3] 杨静 庄蕾[4] 张丹红 陈影 杜向慧 盛李明[2,5] Bai Xue;Yang Jing;Zhuang Lei;Zhang Danhong;Chen Ying;Du Xianghui;Sheng Liming(Department of Radiation Physics,Zhejiang Cancer Hospital,Hangzhou Institute of Medicine(HIM),Chinese Academy of Sciences,Hangzhou 310022,China;Zhejiang Key Laboratory of Thoracic Oncology,Hangzhou 310022,China;Zhejiang Key Laboratory of Radiation Oncology,Hangzhou 310022,China;the Second Clinical Medical College,Zhejiang Chinese Medical University,Hangzhou 310053,China;Department of Thoracic Radiotherapy,Zhejiang Cancer Hospital,Hangzhou Institute of Medicine(HIM),Chinese Academy of Sciences,Hangzhou 310022,China)

机构地区：[1]浙江省肿瘤医院放射物理科,中国科学院杭州医学研究所,杭州310022 [2]浙江省胸部肿瘤重点实验室,杭州310022 [3]浙江省放射肿瘤学重点实验室,杭州310022 [4]浙江中医药大学第二临床研究院,杭州310053 [5]浙江省肿瘤医院胸部放疗科,中国科学院杭州医学研究所,杭州310022

出　　处：《中华放射肿瘤学杂志》2023年第7期620-625,共6页Chinese Journal of Radiation Oncology

基　　金：国家自然科学基金(12005190);北京白求恩基金(flzh202114);浙江省放射肿瘤学重点实验室开放课题(2022ZJCCRAD08)。

摘　　要：目的研究基于剂量组学的局部晚期食管癌根治性放化疗后发生放射性肺炎的影响因素和预测模型。方法回顾性分析2020年1月至2021年8月在浙江省肿瘤医院行根治性放射治疗的105例食管癌病例资料,放射性肺炎的评级按照美国国立癌症研究所常见不良反应术语评定标准5.0版进行评价,分别收集临床因素、传统剂量学特征和剂量组学特征。对用于预测是否发生放射性肺炎的特征进行limma分析。使用支持向量机、k最近邻算法、决策树、随机森林和极致梯度提升算法分别建立预测模型,用十折交叉验证法评估模型的性能,delong检验评估采用不同特征时的模型差异。结果全组患者放射性肺炎发生率为21.9%。1个临床因素、6个传统剂量学特征和42个剂量组学特征与放射性肺炎发生相关(P<0.05)。支持向量机使用线性核函数得到的预测性能最好,未加入和加入剂量组学特征的受试者操作特征曲线下面积分别为0.72和0.75。支持向量机、随机森林和极致梯度提升算法所建立的模型在未加入和加入剂量组学特征时差异有统计学意义(P<0.05)。结论加入剂量组学特征可有效提高食管癌放疗后放射性肺炎的预测模型性能。Objective To study the risk factors and prediction model of radiation pneumonitis(RP)after radical chemoradiotherapy for locally advanced esophageal cancer based on dosiomics.Methods Clinical data of 105 patients with esophageal cancer undergoing radical chemoradiotherapy at Zhejiang Cancer Hospital between January 2020 and August 2021 were retrospectively analyzed.RP was scored using the National Cancer Institute's Common Terminology Criteria for Adverse Events version 5.0(CTCAE 5.0).Clinical factors,traditional dosimetric features and dosiomics features were collected,respectively.The features for predicting PR were analyzed by limma package.Support vector machine,k-nearest neighbor,decision tree,random forest and extreme gradient boosting were used to establish the prediction model,and the ten-fold cross-validation method was employed to evaluate the performance of the model.The differences of this model when different features were chosen were analyzed by delong test.Results The incidence of RP in the whole group was 21.9%.One clinical factor,6 traditional dosimetric features and 42 dosiomics features were significantly correlated with the occurrence of RP(all P<0.05).Support vector machine using linear kernel function yielded the optimal prediction performance,and the area under the receiver operating characteristic(ROC)without and with dosiomics features was 0.72 and 0.75,respectively.The models established by support vector machine,random forest and extreme gradient boosting were significantly different with and without dosiomics features(all P<0.05).Conclusion The addition of dosiomics features can effectively improve the performance of the prediction model of RP after radiotherapy for esophageal cancer.

关 键 词：食管肿瘤 鳞癌 放射疗法 剂量组学 预测模型 

分 类 号：R735.1[医药卫生—肿瘤]