摘要
目的:探讨基于MRI的影像组学模型预测鼻咽癌诱导化疗疗效的价值.方法:回顾性分析经病理证实的184例(132例作为训练集,52例作为验证集)接受诱导化疗的鼻咽癌患者治疗前2周内、诱导化疗结束后T2WI和CE-T1WI两个序列图像.根据实体肿瘤疗效评价标准,将患者分为反应组(102例)和无反应组(82例).分别勾画治疗前两个序列图像中肿瘤的瘤体作为感兴趣区进行影像组学特征提取.分析影像组学特征、临床病理特征及治疗前的血液学指标,构建了 3个模型,用于预测鼻咽癌诱导化疗疗效,分别为模型1(CE-T1WI+T2WI影像组学)、模型2(临床病理特征+治疗前血液学指标)、模型3(模型1+模型2).绘制受试者操作特征(ROC)曲线并评估模型的预测效能.利用决策分析曲线评价不同风险阈值下模型的净获益情况.结果:模型3其训练集的ROC曲线下面积(AUC)值、敏感度、特异度、阳性预测值、阴性预测值分别为0.951、0.831、0.869、0.881、0.815,验证集对应的值分别为0.948、0.774、0.952、0.960、0.741.在训练集及验证集中,模型1的性能均优于模型2(0.940 vs.0.745,0.952 vs.0.608,P 值均＜0.001),同时模型 3 的预测性能均高于模型 2(0.951 vs.0.745,0.948 vs.0.608,P值均＜0.001).决策曲线示三个模型在评价鼻咽癌诱导化疗疗效时均有明确临床获益,并且模型3、模型1均优于模型2.结论:与使用治疗前临床病理特征+血液学指标模型相比,影像组学模型以及影像组学+临床病理+血液学指标模型均能更好地预测鼻咽癌诱导化疗的疗效,对鼻咽癌化疗方案的制定具有一定的指导价值.
Abstract
Objective:To explore the value of MRI radiomic models in the prediction of induction chemotherapy(IC)efficacy in patients with nasopharyngeal carcinoma(NPC).Methods:One hundred and eighty-four patients with pathologically confirmed NPC who received IC were enrolled(132 in the primary set and 52 in the validation set).FS T2WI and CE-TiWI,acquired within 2 weeks before treat-ment and within one week after the end of IC were retrospectively analyzed.Patients were divided into the IC response group(102 cases)and the nonresponse group(82 cases)according to RECIST crite-ria.The primary NPC lesions determined on the pretreatment MRI served as the basis for the volume of interest for feature extraction.In order to predict the efficacy of IC for NPC,radiomic feature,clini-cal pathological feature,and hematological parameters were used to constructed Model 1(CE-T1WI+T2WI radiomic),Model 2(clinical pathological features+pretreatment hematological parameters),and Model 3(Model 1+Model 2),respectively.The discrimination performance of each model was evalua-ted by the receiver operating characteristic curve.Decision curve analysis was conducted to determine the clinical usefulness of the models by quantifying the net benefits at different threshold probabilities.Results:The AUC,sensitivity,specificity,positive predictive value,and negative predictive value of Model 3 were 0.951,0.831,0.869,0.881,and 0.815 in the primary set,and 0.948,0.774,0.952,0.960,and 0.741 in the validation set,respectively.The performance of the Model 1 was better than the Model 2 in the both cohorts(0.940 vs.0.745,0.952 vs.0.608,both P＜0.001),and the performance of the Model 3 was also better than the Model 2 in the both cohorts(0.951 vs.0.745,0.948 vs.0.608,both P＜0.001).The Model 3 and Model 1 were more beneficial than Model 2 in clinical practice,which il-lustrated by decision curve analysis.Conclusion:Compared with the Model 2,Model 1 and Model 3 can better predict the efficacy of IC for NPC,which may aid clinical decision-making and potentially guide the chemotherapy regimen for NPC.
基金项目
四川省科技计划基金资助项目(2021YFG0125)
四川省科技计划基金资助项目(2022YFSY0006)
NETL
NSTL
万方数据