摘要
目的:基于深度学习算法,使用临床数据构建人工智能(artificial intelligence,AI)模型,评估其预测非小细胞肺癌(non-small cell lung cancer,NSCLC)新辅助免疫联合化疗疗效的可行性.方法:收集 2020年 1月至 2024年 1月于首都医科大学附属北京胸科医院确诊且接受新辅助免疫联合化疗的 132例NSCLC患者的临床病理数据.对影响新辅助免疫联合化疗疗效的主要因素进行统计学分析.根据统计学分析结果以及查阅相关文献对变量进行筛选,构建变量数据集.基于多层感知机(multi-layer perceptron,MLP)算法采用 5折交叉验证建立深度学习模型,应用接收者操作特征(receiver operating characteristic,ROC)曲线评估模型性能.结果:132例患者中单因素分析显示获得主要病理缓解(major pathological response,MPR)组与未获得MPR组的患者在性别(P=0.020)、吸烟史(P=0.004)、癌胚抗原(carcinoembryonic antigen,CEA)(P=0.038)及程序性死亡配体-1(programmed death-ligand 1,PD-L1)是否≥1%(P=0.038)方面差异具有统计学意义;获得完全病理缓解(complete pathologic-al response,pCR)组与未获pCR组患者在肿瘤大小是否≤3 cm(P=0.007)和CEA水平方面(P=0.010)差异具有统计学意义.经 5折交叉验证后,MPR预测模型在验证集和测试集的平均受试者工作特征曲线下面积(area under the curve,AUC)值分别为0.72及0.71.结论:该深度学习模型能够有效预测NSCLC患者新辅助免疫联合化疗疗效.
Abstract
Objective:An artificial intelligence(AI)model based on deep learning algorithms was constructed using clinical data to evaluate the feasibility of predicting the efficacy of neoadjuvant immunotherapy combined with chemotherapy for non-small cell lung cancer(NSCLC).Methods:Clinical and pathological data of 132 patients with NSCLC who were diagnosed and treated with neoadjuvant immunotherapy combined with chemotherapy between January 2020 and January 2024 at Beijing Chest Hospital/Beijing Tuberculosis and Thoracic Tumor Research Institute were collected.Statistical analysis was conducted to identify the main factors affecting the efficacy of neoadjuvant im-munotherapy combined with chemotherapy.Variables were selected based on statistical results and relevant literature,and a variable data-set was constructed.A deep learning model was established using a multi-layer perceptron(MLP)algorithm with 5-fold cross-validation,and the performance of the model was evaluated using receiver operating characteristic curve(ROC).Results:Among the 132 patients,univari-ate analysis demonstrated statistically significant differences in sex(P=0.020),smoking history(P=0.004),carcinoembryonic antigen(CEA)(P=0.038)and programmed death-ligand 1(PD-L1)≥1%(P=0.038)between the major pathological response(MPR)and non-MPR groups.Patients in the complete pathological response(pCR)group and non-pCR groups showed statistical differences in tumor size(P=0.007)and CEA levels(P=0.010).After 5-fold cross-validation,the average area under the curve(AUC)of the MPR prediction model in the validation and test sets was 0.72 and 0.71,respectively.Conclusions:The deep learning model can effectively predict the efficacy of neoadjuvant chemoim-munotherapy in patients with NSCLC.
维普
万方数据