基于CT图像的自动编码器人工智能方法预测食管癌术前化疗联合免疫治疗的疗效  

Efficacy prediction on preoperative chemotherapy combined with immunotherapy for esophageal carcinoma using an autoencoder-based artificial intelligence approach on CT images

作　　者：程蕾[1] 沈海霞 蒋友华[2] 陈奇勋[2] 闫萌 陶开义[2] CHENG Lei;SHEN Haixia;JIANG Youhua;CHEN Qixun;YAN Meng;TAO Kaiyi(Department of Thoracic Radiotherapy,Zhejiang Cancer Hospital,Hangzhou 310022,China;不详)

机构地区：[1]浙江省肿瘤医院胸部放疗科,杭州310022 [2]浙江省肿瘤医院食管外科,杭州310022 [3]温州医科大学研究生培养基地 [4]天津医科大学肿瘤医院放射治疗科

出　　处：《浙江医学》2025年第4期348-355,共8页Zhejiang Medical Journal

基　　金：浙江省医药卫生科技计划项目(2022KY623);浙江省中医药科技计划项目(2024ZL022)。

摘　　要：目的建立一种基于CT图像的自动编码器人工智能方法以预测食管癌术前化疗联合免疫治疗的疗效。方法回顾性分析2019年7月至2023年7月在浙江省肿瘤医院接受新辅助化疗联合免疫治疗后行手术治疗的240例食管鳞状细胞癌患者。提取患者临床病理信息,并根据肿瘤病理消退情况,将所有患者分为病理完全缓解组(35例)和非病理完全缓解组(205例)。基于CT图像的自动编码器人工智能方法,利用变分自动编码器提取任务相关“深度学习特征”,将患者按4∶1的比例随机分为训练集192例和验证集48例,建立病理完全缓解分类模型,并采用5折交叉验证分组,最终采用多个指标评估模型的预测效能:精确率、召回率、F1分数、AUC、准确率。结果训练集的精确率为0.665,召回率为0.888,F1分数为0.760,AUC为0.946,准确率为0.901;验证集的精确率为0.651,召回率为0.836,F1分数为0.726,AUC为0.935,准确率为0.896。结论基于CT图像的自动编码器人工智能方法可以有效地预测食管癌新辅助化疗联合免疫治疗的疗效,为患者制定个体化治疗提供依据。Objective To establish a novel deep learning model based on CT images for predicting the efficacy of neoadjuvant therapy.Methods A retrospective analysis was conducted on 240 patients with esophageal squamous cell carcinoma who underwent neoadjuvant chemotherapy combined with immunotherapy followed by surgical treatment at Zhejiang Cancer Hospital from July 2019 to July 2023.Clinical and pathological information of patients was extracted,and all patients were divided into two groups based on pathological response:pathological complete response(pCR,n=35)and non-pathological complete response(non-pCR,n=205).By using deep learning methods,a variational autoencoder was employed to extract taskrelated"deep learning features",the patients were randomly allocated to a training set(n=192)and a validation set(n=48)in a 4∶1 ratio,and a classification model for pathological complete response was established.The model was evaluated using 5-fold cross-validation,and its predictive performance was assessed using multiple metrics:precision rate,recall rate,F1 score,AUC,and accuracy rate.Results In the training set,the precision rate was 0.665,recall rate was 0.888,F1 score was 0.760,AUC was 0.946,and accuracy was 0.901.In the validation set,the precision rate was 0.651,recall rate was 0.836,F1 score was 0.726,AUC was 0.935,and accuracy rate was 0.896.Conclusion The autoencoder-based artificial intelligence approach using CT images can effectively predict the therapeutic efficacy of neoadjuvant immunotherapy combined with chemotherapy for esophageal squamous cell carcinoma,providing a basis for individualized treatment regimen for patients.

关 键 词：食管癌 新辅助治疗 免疫治疗 深度学习 病理完全缓解 

分 类 号：R73[医药卫生—肿瘤]