体素内不相干运动弥散加权成像影像组学列线图预测高级别脑胶质瘤IDH1基因突变  

Prediction of IDH1 Gene Mutation Status in High-grade Brain Gliomas Using a Radiomics Nomogram Based on Intravoxel Incoherent Motion Diffusion-Weighted Imaging

作　　者：刘梦茹 董海波 林雪瑶 郑金晶 LIU Mengru;DONG Haibo;LIN Xueyao;ZHENG Jinjing(Department of Radiology,The Affiliated Lihuili Hospital,Ningbo University,Ningbo 315041,China)

机构地区：[1]宁波大学附属李惠利医院放射科,宁波315041

出　　处：《中国医学计算机成像杂志》2025年第1期20-26,共7页Chinese Computed Medical Imaging

基　　金：浙江省医药卫生科技项目(2023KY1047)。

摘　　要：目的:探讨体素内不相干运动弥散加权成像(IVIM-DWI)影像组学列线图预测高级别脑胶质瘤异柠檬酸脱氢酶1(IDH1)基因突变的价值。方法:回顾性收集2017年1月至2024年3月收治的94例术后病理证实为高级别脑胶质瘤患者的常规MR和IVIM-DWI资料,其中,34例为IDH1基因突变型(IDH1-M),60例为IDH1基因野生型(IDH1-W),以7∶3比例随机划分为训练集(66例)和测试集(28例),使用3D-Slicer软件基于常规MRI图像勾画肿瘤实质区、坏死区、水肿区作为感兴趣区,分别投影到D、D^(*)、f序列,每个感兴趣区中提取出1037个影像组学特征,每例患者共计3111个特征。采用最小冗余最大相关性(mRMR)及最小绝对收缩和选择算子(LASSO)进行特征筛选,建立logistic回归影像组学模型,计算Radscore。纳入筛选后的临床特征,构建临床-影像组学联合模型,生成列线图。分别建立D、D^(*)、f 3个参数模型。采用受试者工作特征(ROC)曲线评价预测模型的效能,以临床决策曲线分析(DCA)评估列线图的实用性。结果:经筛选后将年龄及强化程度这2个因素结合影像组学标签构建了3个参数预测模型。D参数模型训练集及测试集ROC曲线下面积(AUC)值分别为0.94、0.92,D^(*)参数模型AUC值分别为0.84、0.81,f参数模型AUC值分别为0.90、0.85。DCA结果表明,3个参数模型的净收益均较高,D模型的净收益在几乎整个Pt值范围内高于D^(*)及f模型。结论:基于IVIM-DWI影像组学列线图能有效预测高级别脑胶质瘤IDH1基因突变状态,以D参数模型最优。Purpose:To investigate the predictive value based on intravoxel incoherent motion diffusionweighted imaging(IVIM-DWI)radiomics nomogram for isocitrate dehydrogenase 1(IDH1)gene mutations in highgrade gliomas.Methods:We retrospectively analyzed the routine MR and IVIM-DWI data of 94 patients with surgically and pathologically confirmed high-grade brain gliomas from January 2017 to March 2024.The data of 34 patients with mutant-type IDH1(IDH1-M)and 60 patients with wild-type IDH1(IDH1-W)were randomly divided into training(N=66)and test sets(N=28)at a 7∶3 ratio.We used the 3D Slicer software to delineate the tumor parenchyma,necrosis,and edema areas as regions of interest(ROIs)based on conventional MRI images and projected them onto the D,D^(*)and f sequence images.We extracted 1037 radiomic features from each ROI,with a total of 3111 features per patient.The minimum redundancy-maximum relevance(mRMR)algorithm and least absolute shrinkage and selection operator(LASSO)were used to select the optimal features,and a logistic regression(LR)radiomics model was established and the Radscore was calculated.The screened clinical features were incorporated to construct a combined clinical-radiomics model,and a nomogram was generated.Three separate models were established for the parameters D,D^(*),and f.Receiver operating characteristic(ROC)curve was used to evaluate the efficacy of the prediction model,and clinical decision curve analysis(DCA)was used to evaluate the practicability of the nomogram.Results:After selection,age and enhancement level were combined with radiomic features to build three separate predictive models.The area under ROC curve(AUC)values for the training and validation sets of the D parameter model were 0.94 and 0.92,respectively.The AUC values of the D^(*)parameter model were 0.84 and 0.81,respectively,and the AUC values of the f parameter model were 0.90 and 0.85,respectively.Decision curve analysis indicated that all three parameter models were with high net benefits,with the net benefit of the D m

关 键 词：脑胶质瘤 异柠檬酸脱氢酶1 体素内不相干运动 弥散加权成像 列线图 

分 类 号：R445.2[医药卫生—影像医学与核医学]