静脉期CT影像组学预测新辅助化疗用于局部进展期胃癌效果  

Venous CT radiomics for predicting effect of neoadjuvant chemotherapy for locally advanced gastric cancer

作　　者：韩晓梦 刘顺利 林吉征[1] 娄和南 宋洪政 王博[1] 宋瑶琳 赵晓丹[1] HAN Xiaomeng;LIU Shunli;LIN Jizheng;LOU Henan;SONG Hongzheng;WANG Bo;SONG Yaolin;ZHAO Xiaodan(Department of Radiology,the Affiliated Hospital of Qingdao University,Qingdao 266003,China;Department of Pathology,the Affiliated Hospital of Qingdao University,Qingdao 266003,China)

机构地区：[1]青岛大学附属医院放射科,山东青岛266003 [2]青岛大学附属医院病理科,山东青岛266003

出　　处：《中国介入影像与治疗学》2025年第1期37-42,共6页Chinese Journal of Interventional Imaging and Therapy

摘　　要：目的观察静脉期CT影像组学预测新辅助化疗(NACT)用于局部进展期胃癌(LAGC)效果的价值。方法回顾性收集接受NACT的325例LAGC患者,以247例为训练集、78例为验证集。根据术后病理所示肿瘤退缩分级(TRG)评价NACT疗效。以单因素logistic回归分析并筛选临床指标,构建预测NACT治疗LAGC效果的临床模型。分别于NACT前、后增强静脉期CT中提取病灶影像组学特征,计算Delta影像组学特征(即NACT前、后影像组学特征差值与前者的比值)并筛选最佳者构建影像组学标签;筛选最优标签,以之联合临床模型构建联合模型。绘制受试者工作特征(ROC)曲线,计算曲线下面积(AUC),评价各模型预测效能;以决策曲线分析(DCA)评估各模型临床价值。结果训练集67例疗效显著、180例疗效不显著;验证集18例疗效显著、60例疗效不显著。LAGC Borrmann分型为NACT疗效的预测因素(P=0.031),以之构建的临床模型在训练集和验证集的AUC分别为0.577和0.520。基于NACT前、后及Delta影像组学特征分别筛选出19、14及17个最佳特征,以之建立的NACT前(Pre-Rad)、后(Post-Rad)及Delta(Delta-Rad)影像组学标签在训练集的AUC分别为0.672、0.796及0.789,在验证集分别为0.558、0.805及0.666,其中Post-Rad最佳;以之联合临床模型构建的联合模型在训练集和验证集的AUC分别为0.824和0.818,均高于临床模型(P均<0.001)而与Post-Rad差异无统计学意义(P均>0.05);阈值为0.4~0.7时,联合模型临床净获益高于临床模型及Post-Rad。结论静脉期CT影像组学可有效预测NACT治疗LAGC效果;联合临床特征可提高预测效能。Objective To investigate the value of CT radiomics for predicting effect of neoadjuvant chemotherapy(NACT)for locally advanced gastric cancer(LAGC).Methods Totally 325 LAGC patients who received NACT were retrospectively enrolled,among them 247 were taken as training set,while the rest 78 were taken as validation set.Tumor regression scale(TRG)was evaluated according to postoperation pathology after NACT,and the efficacy of NACT was evaluated.Univariate logistic regression was used to analyze and screen clinical predictors of effect of NACT,and clinical model was constructed.Radiomics features were extracted based on venous phase enhanced CT pre-and post-NACT,and Delta radiomics features(i.e.the ratio of the difference of pre-and post-NACT radiomics features and pre-NACT radiomics features)were calculated.The best features were screened based on pre-NACT,post-NACT and Delta radiomics features to construct radiomics labels,the optimal label was screened and used to construct combined model through combining clinical model.Receiver operating characteristic(ROC)curve was plotted,and the area under the curve(AUC)was calculated to evaluate predicting efficiency of the above models.Decision curve analysis(DCA)was performed to explore the clinical value of each model.Results In training set,significant effect was found in 67 cases,but not in 180 cases,while in validation set,significant effect was found in 18 cases but not in 60 cases.Borrmann classification of LAGC before NACT was the clinical predictor(P=0.031),and clinical model was constructed,which had AUC of 0.577 and 0.520 in training and validation sets,respectively.Based on pre-NACT,post-NACT and Delta radiomics features,19,14 and 17 best features were selected,and AUC of the established radiomics labels of Pre-Rad,Post-Rad and Delta-Rad in training set was 0.672,0.796 and 0.789,while in validation set was 0.558,0.805 and 0.666,respectively.Post-Rad was the optimal label,which was used to construct combined model.AUC of the obtained combined model in training a

关 键 词：胃肿瘤 体层摄影术 X线计算机 新辅助化疗 治疗转归 影像组学 

分 类 号：R735.2[医药卫生—肿瘤] R814.42[医药卫生—临床医学]