基于机器学习算法构建胸段食管鳞癌患者化疗联合免疫治疗反应预测模型的研究  

Development of a prediction model for chemotherapy and immunotherapy response in esophageal squamous cell carcinoma patients using machine learning algorithms

作　　者：陈金城 张小勤 刘杰 李童心 吴毅[2] 何萍 吴蔚 CHEN Jincheng;ZHANG Xiaoqin;LIU Jie;LI Tongxin;WU Yi;HE Ping;WU Wei(Department of Thoracic Surgery,First Affiliated Hospital,Army Medical University(Third Military Medical University),Chongqing;Department of Digital Medicine,College of Biomedical Engineering and Medical Imaging,Army Medical University(Third Military Medical University),Chongqing,China;Department of Cardiac Surgery,First Affiliated Hospital,Army Medical University(Third Military Medical University),Chongqing)

机构地区：[1]陆军军医大学(第三军医大学)第一附属医院胸外科,重庆 [2]陆军军医大学(第三军医大学)生物医学工程与医学影像学院数字医学教研室,重庆 [3]陆军军医大学(第三军医大学)第一附属医院心脏外科,重庆

出　　处：《陆军军医大学学报》2025年第6期591-601,共11页Journal of Army Medical University

基　　金：重庆市科卫联合医学科研项目(2022ZDXM018)。

摘　　要：目的应用不同机器学习算法构建食管鳞癌患者化疗联合免疫治疗反应预测模型,并筛选出最优模型。方法回顾性收集2022年1月至2023年12月在陆军军医大学第一附属医院胸外科住院,行化疗联合免疫治疗的174例食管鳞癌患者,收集患者治疗前的CT及临床信息。采用简单随机抽样法将患者按7:3比例随机分为训练集(n=122)与测试集(n=52)。提取并筛选CT影像组学特征,使用5种机器学习算法构建影像组学模型和临床-影像组学模型。在训练集中进行5折交叉验证,在测试集中采用受试者工作特征曲线(receiver operating characteristic curve,ROC)、F1分数评估预测模型性能,对于表现最佳模型使用局部可解释模型不可知的解释(local interpretable model-agnostic explanations,LIME)算法进行解释。结果174例患者中有115例患者出现临床缓解(66.1%)。从患者临床信息与CT影像中筛选出1个临床特征、10个影像组学特征。基于5种机器学习算法构建的影像组学模型和临床-影像组学模型的最佳ROC曲线下面积(area under of receiver operating characteristic,AUC)分别是0.750(95%CI:0.616~0.883)和0.766(95%CI:0.647~0.895)。最优临床-影像组学模型的F1分数为0.829。基于LIME算法,最佳模型对实例样本的预测显示出可靠性。结论基于机器学习算法构建的临床-影像组学预测模型性能良好,通过预测食管鳞癌患者化疗联合免疫治疗反应,为医生临床决策提供参考。Objective To develop models for predicting response to chemotherapy combined with immunotherapy in patients with esophageal squamous carcinoma with various machine learning algorithms,and then select the optimal model.Methods A retrospective study was performed for 174 patients with esophageal squamous cell carcinoma undergoing chemotherapy combined with immunotherapy admitted in Department of Thoracic Surgery of the First Affiliated Hospital of Army Medical University from January 2022 to December 2023.The CT scans and clinical information were collected before treatment.They were randomly divided into a training set(n=122)and a testing set(n=52)in a ratio of 7∶3.CT radiomic features were extracted and selected,and then 5 machine-learning algorithms were employed to establish the prediction models,including radiomics model and clinical-radiomics model.Five-fold cross-validation was conducted on the training set,and the performance of the prediction models was evaluated on the testing set using receiver operating characteristic(ROC)curve and the F1 score.The best-performing model was further explained using local interpretable model-agnostic explanations(LIME)algorithm.Results Among the 174 patients,115(66.1%)achieved clinical remission.From the clinical information and CT images,1 clinical features and 10 radiomic features were identified.The area under of ROC curve(AUC)for the radiomics and clinical-radiomics models was 0.750(95%CI:0.616~0.883),and 0.766(95%CI:0.637~0.895),respectively.The F1 score of the optimal clinical-radiomics model was 0.829.LIME algorithm indicated that this best model demonstrated reliability in predicting individual samples.Conclusion The clinical-radiomics prediction model based on machine learning algorithm performs well,and can provide a reference for doctors'clinical decision-making by predicting the response to chemotherapy combined with immunotherapy in patients with esophageal squamous cell carcinoma.

关 键 词：食管癌 机器学习 影像组学 

分 类 号：R319[医药卫生—基础医学] R730.5R735.1