多参数MRI影像组学模型对进展期鼻咽癌局部复发的预测价值  

Predictive value of multi-parametric MRI radiomics model for local recurrence of advanced nasopharyngeal carcinoma

作　　者：倪云燕 陈照 陈韵彬[1] NI Yunyan;CHEN Zhao;CHEN Yunbin(Department of Diagnostic Radiology,Fujian Cancer Hospital&Fujian Medical University Cancer Hospital,Fuzhou 350000,Fujian Province,China;Department of Radiology,Fuzhou Pulmonary Hospital,Fuzhou 350000,Fujian Province,China)

机构地区：[1]福建省肿瘤医院,福建医科大学附属肿瘤医院放诊科,福建福州350000 [2]福建省福州肺科医院影像科,福建福州350000

出　　处：《肿瘤影像学》2022年第1期57-63,共7页Oncoradiology

摘　　要：目的:探讨多参数磁共振成像(magnetic resonance imaging,MRI)影像组学模型对进展期鼻咽癌(nasopharyngeal carcinoma,NPC)局部复发的预测价值。方法:回顾并分析经病理学检查证实的86例进展期NPC的临床及影像学资料,将35例经病理学检查证实复发的患者列入复发组,51例随访无复发的患者列入无复发组。所有患者均行MRI扫描,采集初诊、放疗结束时及放疗后6个月共3个时间点的轴位T1加权成像(T1-weighted imaging,T1WI)增强扫描、T2加权成像(T2 weighted imaging,T2WI)-短时间反转恢复(short time inversion recovery,STIR),采用Spearman相关分析提取特征参数,回归特征消除(recursive feature elimination,RFE)算法对冗余特征进行降维,再通过Ridge Classifier分类器学习得到预测模型,5折交叉验证法训练并验证模型。结果:进展期NPC复发组的复发间隔时间为7~61个月,中位时间为24个月。基于3个时间点T2WI-STIR及T1WI增强图像最终提取了15个与进展期NPC局部复发相关的特征参数,其中包括灰度共生矩阵(gray level co-occurrence matrix,GLCM)参数8个,灰度游程矩阵(gray level run length matrix,GLRLM)参数3个,FirstOrder特征集参数2个,相邻灰度差分矩阵(neighboring gray tone difference matrix,NGTDM)参数1个,灰度级带矩阵(gray level size zone matrix,GLSZM)参数1个;初诊时图像特征参数2个,放疗结束时9个,放疗后4个。依据特征参数建立线性回归模型,其平均灵敏度为82.9%,平均特异度为98.0%,平均准确度为91.8%,平均曲线下面积(area under curve,AUC)值为0.953。结论:多参数MRI放射组学模型对预测进展期NPC患者局部复发的性能较好,能为评估进展期NPC复发风险提供参考。Objective:To explore the predictive value of multi-parametric magnetic resonance imaging(MRI)radiomics model for local recurrence of advanced nasopharyngeal carcinoma(NPC).Methods:A total of 86 cases pathologically confirmed advanced NPC with clinical data and imaging data were analyzed retrospectively.According to the follow-up data,35 patients of NPC recurrence confirmed by pathology were included in recurrent group.The rest of 51 patients without recurrent evidence at pathology and imaging data were included in non-recurrent group.All enrolled subjects underwent MRI scans.Contrast-enhanced T1-weighted imaging(T1WI),T2 weighted imaging(T2WI)-short time inversion recovery(STIR)were acquired at the initial diagnosis,the end of radiotherapy and 6 months after radiotherapy respectively.Spearman correlation analysis and recursive feature elimination(RFE)were used to extract and reduce the redundant features.The predictive model is constructed by the Ridge Classifier learning.Additionally,the model was trained and verified by the 5-fold cross-validation method.Results:The recurrence interval was 7-61 months,and the median time was 24 months.Based on three time points of T2WI-STIR and contrast-enhanced T1WI images,15 features parameters related to local recurrence of advanced NPC were extracted,including 8 parameters of gray level co-occurrence matrix(GLCM),3 parameters of gray level run length matrix(GLRLM),2 parameters of FirstOrder feature set,1 parameter of neighboring gray tone difference matrix(NGTDM)and 1 parameter of gray level size zone matrix(GLSZM);including 2 feature parameters at initial diagnosis,9 at the end of radiotherapy and 4 after radiotherapy.A linear regression model was obtained.The average sensitivity,specificity,accuracy and area under curve(AUC)were 82.9%,98.0%,91.8%and 0.953,respectively.Conclusion:Multi-parametric MRI radiomics model has a good performance in predicting the local recurrence of advanced NPC,which can provide a reference for evaluating the risk of recurrent NPC.

关 键 词：鼻咽癌 磁共振成像 影像组学 复发 预测模型 

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