基于多参数MRI影像组学的列线图预测鼻咽癌诱导化疗效果  

Nomogram 4Based on Multiparameter MRI Radiomics in Predicting the Efficacy of Induction Chemotherapy in Nasopharyngeal Carcinoma

作　　者：王卓 张少茹 周云舒 张若弟 刘世莉 丁伟[3] 张自新[3] 陈志强 WANG Zhuo;ZHANG Shaoru;ZHOU Yunshu;ZHANG Ruodi;LIU Shili;DING Wei;ZHANG Zixin;CHEN Zhiqiang(Department of Radiology,the First Affiliated Hospital of Hainan Medical University,Haikou 570102,China;不详;Department of Radiology,General Hospital of Ningxia Medical University,Yinchuan 750004,China)

机构地区：[1]海南医学院第一附属医院放射科,海南海口570102 [2]宁夏医科大学临床医学院,宁夏银川750004 [3]宁夏医科大学总医院放疗科,宁夏银川750004 [4]宁夏医科大学总医院放射科,宁夏银川750004

出　　处：《中国医学影像学杂志》2023年第5期459-466,共8页Chinese Journal of Medical Imaging

基　　金：宁夏回族自治区重点研发计划(2019BEG03033);宁夏自然科学基金(2022AAC03472)。

摘　　要：目的 探讨基于多参数MRI的影像组学结合临床因素及MRI强化程度的列线图预测局部晚期鼻咽癌患者诱导化疗疗效的价值。资料与方法 回顾性分析2014年7月—2022年4月宁夏医科大学总医院184例Ⅲ、Ⅳ期局部晚期鼻咽癌,按照3∶2随机分为训练组(n=110)和验证组(n=74)。用3D-Slicer勾画感兴趣区并用Pyradiomics包提取特征。使用多因素Logistic回归选择诱导化疗疗效的临床预测因子。采用最小绝对收缩和选择算法(LASSO)筛选特征,通过多变量Logistic回归分别构建临床、影像组学模型及联合模型,并绘制列线图。以受试者工作特征曲线下面积(AUC)和Delong检验评估和比较3种模型的预测效能。应用决策曲线分析观察列线图的临床净获益。结果 通过Logistic回归分析纳入2个临床预测因子,包括T分期(OR=0.335,P=0.001)、癌灶MRI强化程度(OR=5.177,P=0.003)。通过LASSO-Logistic回归分别从CE_T1WI_FS、T1WI、T2WI_FS中筛选出2、7、6个与化疗敏感度相关的组学特征并计算影像组学评分。与临床、影像组学模型比较,联合模型预测效能最佳(训练组AUC:0.922比0.748、0.851,Z=3.682、2.680,P<0.01;验证组AUC:0.918比0.782、0.843,Z=3.073、2.409,P<0.05)。决策曲线分析显示,当阈值在0.20~0.85时,联合模型的临床净获益水平高于单一临床或影像组学模型。结论 基于治疗前多参数MRI的影像组学评分、T分期和癌灶MRI强化程度是诱导化疗疗效的独立预测因子,三者联合可以提高预测效能,为局部晚期鼻咽癌患者的个性化治疗提供依据。Purpose To investigate the performance of nomogram based on multiparametric MRI combined with clinical factors and degree of enhancement on MRI for predicting the response to induction chemotherapy in patients with loco-regionally advanced nasopharyngeal carcinoma.Materials and Methods 184 patients with loco-regionally advanced nasopharyngeal carcinoma(stageШandⅣ)in the General Hospital of Ningxia Medical University from July 2014 to April 2022 were retrospectively analyzed.All patients were randomly stratified into training(n=110)and testing(n=74)cohorts at 3∶2 ratios.The region of interest was segmented via 3D-Slicer and features were extracted via Pyradiomics package.Multivariate Logistic regression analysis was applied to identify the predictive clinical factors.Least absolute shrinkage and selection operator(LASSO)was performed to select features.The clinical,radiomics and combined models were established by multivariate logistic regression,and the nomogram was finally obtained.The performance of the three models was assessed by area under the curve(AUC)and Delong test,respectively.Decision curve analysis was performed to evaluate the net benefit of the nomogram.Results Two predictive clinical factors were identified via logistic regression analysis,including T stage(OR=0.335,P=0.001),enhancement degree of the foci(OR=5.177,P=0.003).Two,seven and six radiomics features of chemotherapy sensitivity were selected from CE_T1WI_FS,T1WI and T2WI_FS via LASSO-Logistic regression,respectively;and Radiomics score was further calculated.Compared with clinical or radiomics models,nomogram model showed the best predictive performance(AUC for training group:0.922 vs.0.748,0.851,Z=3.682,2.680,P<0.01;AUC for testing group:0.918 vs.0.782,0.843,Z=3.073,2.409,P<0.05).Decision curve analysis displayed that when the threshold was in the range of 0.20 to 0.85,the level of net clinical benefit of the combined model was significantly higher than that of the single clinical or radiomics model.Conclusion Radiomics score based o

关 键 词：鼻咽癌 磁共振成像 影像组学 诱导化疗 列线图表 

分 类 号：R739.63[医药卫生—肿瘤] R445.2[医药卫生—临床医学]