基于多参数MRI影像组学融合模型对较低级别脑胶质瘤患者生存预后的术前预测价值  

Preoperative predictive value of multiparameter MRI radiomics fusion model for prognosis of low-grade glioma

作　　者：马平[1] 李文菲[2] 王艺璇 李淑雅 杨哲 伍建林[4] 李彦国[2] 韩芳 Ma Ping;Li Wenfei;Wang Yixuan;Li Shuya;Yang Zhe;Wu Jianlin;Li Yanguo;Han Fang(Department of Radiology,the Second Hospital of Qinhuangdao,Qinhuangdao 066000,China;Department of Radiology,the First Hospital of Qinhuangdao,Qinhuangdao 066000,China;Department of PET/CT Medical Center,Affiliated Zhongshan Hospital of Dalian University,Dalian 116001,China;Department of Radiology,Affiliated Zhongshan Hospital of Dalian University,Dalian 116001,China;Department of Radiology,Affiliated Xinhua Hospital of Dalian University/Dalian Key Laboratory of Functional and Molecular Imaging,Dalian 116001,China)

机构地区：[1]秦皇岛市第二医院影像科,秦皇岛066000 [2]秦皇岛市第一医院影像科,秦皇岛066000 [3]大连大学附属中山医院PET/CT医学中心,大连116001 [4]大连大学附属中山医院影像科,大连116001 [5]大连大学附属新华医院影像科/大连市功能分子影像重点实验室,大连116001

出　　处：《中华解剖与临床杂志》2025年第3期154-162,共9页Chinese Journal of Anatomy and Clinics

基　　金：国家自然科学基金项目(82071911);辽宁省重点研发计划(2022JH2/101300021);秦皇岛市科技项目(202301A279)。

摘　　要：目的探讨多参数MRI影像组学联合临床病理特征的融合模型对较低级别脑胶质瘤(LGGs)患者生存预后的术前预测价值。方法回顾性队列研究。纳入癌症基因图谱公共数据库中1991年1月—2001年5月105例WHO肿瘤分级为Ⅱ~Ⅲ级的LGGs患者为训练集,其中男53例、女52例,年龄42(33,57)岁。纳入2017年1月—2021年11月秦皇岛市第二医院52例LGGs患者为外部验证集,其中男26例、女26例,年龄42(31,54)岁。记录所有患者临床病理特征(性别、年龄、肿瘤分级、IDH基因1p/19q状态、生存状态及生存时间)和MRI影像资料。对每例患者的T1加权像(WI)、T2WI、T2液体衰减反转恢复序列及增强T1WI图像进行特征提取,在训练集中采用组内相关系数、方差阈值法、Spearman相关分析、单因素Cox回归分析、最小绝对收缩与选择算子算法(LASSO)、LASSO-Cox回归分析进行图像特征筛选,构建影像组学风险评分(RRS)模型。根据训练集患者生存时间采用X-tile软件确定RRS模型预测3年累积生存率的最佳临界值,将LGGs患者分为高风险组和低风险组,采用Kaplan-Meier曲线评估训练集中高、低风险组患者生存率的差异,并在验证集中进行验证。在训练集中对高风险组和低风险组LGGs患者年龄、性别、异柠檬酸脱氢酶(IDH)基因1p/19q突变状态及肿瘤分级等临床病理特征行单因素、多因素Cox回归分析,选出影响LGGs生存预后的独立危险因素,联合影像组学特征构建预测LGGs患者术后3年累积生存率的融合模型。在训练集中采用时间依赖受试者操作特征曲线(ROC曲线)评估RRS模型和融合模型对LGGs患者3年累积生存率的预测效能,采用Delong检验比较不同模型的预测效能,通过校正曲线评估融合模型的校正效能,并在验证集中进行验证。结果训练集与验证集患者的性别、年龄、肿瘤分级、IDH基因1p19q状态比较,差异均无统计学意义(P值均>0.05)。MRI图像经�Objective This study aimed to explore the value of the fusion model of multiparameter MRI combined with clinicopathological features in predicting the survival and prognosis of lower-grade gliomas(LGGs).Methods A retrospective cohort study was conducted.The training set included 105 LGG patients with tumor gradeⅡ-Ⅲ(53 males,52 females,age 42[33,57]years)from the Cancer Genome Atlas public database,with diagnoses made between January 1991 and May 2001.The external validation set included 52 LGG patients(26 males,26 females,age 42[31,54]years)from Qinhuangdao Second Hospital,recruited between January 2017 and November 2021.Clinical and pathological data(gender,age,tumor grade,IDH gene 1p/19q status,survival status,and survival time)and MRI imaging data were recorded for all patients.Features were extracted from each patient's T1 weighted images(WI),T2WI,T2 fluid-attenuated inversion recovery,and contrast-enhanced T1WI images.In the training set,image feature selection was performed using intra-class correlation coefficient,variance threshold method,Spearman correlation analysis,univariate Cox regression,least absolute shrinkage and selection operator(LASSO),and LASSO-Cox regression analysis to build a radiomics risk score(RRS)model.The optimal cutoff value of the RRS model for predicting the three-year cumulative survival rate was determined using X-tile software on the basis of the survival time of the training set patients,categorizing LGG patients into high-risk and low-risk groups.Kaplan-Meier curves were used to assess the difference in survival between the highand low-risk groups in the training set and were validated in the validation set.In the training set,univariate and multivariate Cox regression analyses of clinical-pathological features,including age,gender,IDH gene mutation status,and tumor grade,were conducted to identify independent prognostic factors for LGG survival.A fusion model was then developed by integrating radiomics features with clinical-pathological data to predict the three-year cum

关 键 词：神经胶质瘤 预后 较低级别胶质瘤 磁共振成像 影像组学 预测模型 

分 类 号：R739.41[医药卫生—肿瘤]