基于MRI灌注成像的全脑及局部影像组学特征预测急性脑卒中预后的研究  

The study of MRI perfusion imaging global and local radiomics in the prediction of outcome in acute stroke

作　　者：周啸虎[1] 彭明洋 王同兴[1] 陈国中 殷信道[1] 任军[1] ZHOU Xiaohu;PENG Mingyang;WANG Tongxing;CHEN Guozhong;YIN Xindao;REN Jun(Department of Medical Imaging,Nanjing First Hospital,Nanjing Medical University,Nanjing 210006,China)

机构地区：[1]南京医科大学附属南京医院(南京市第一医院)医学影像科,南京210006

出　　处：《磁共振成像》2024年第10期98-102,135,共6页Chinese Journal of Magnetic Resonance Imaging

基　　金：国家自然科学基金项目(编号:82001811);江苏省重点研发计划项目(编号:BE2021604)。

摘　　要：目的探讨MR灌注加权成像(perfusion weighted imaging,PWI)的局部特征和全脑特征在急性脑卒中血管内治疗后预后中的价值。材料与方法回顾性分析在我院就诊的180例急性脑卒中患者的PWI图像。采用ITK-SNAP软件勾画Tmax图的灌注异常感兴趣区,应用SPM软件自动分割Tmax图全脑区。应用AK软件分别提取局部和全脑特征并降维,通过支持向量机分类器构建急性脑卒中预后模型并寻找最优预测模型。结果经最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)降维后与预后最相关的局部特征为6个、全脑特征为5个、联合局部和全脑的特征为10个。受试者工作特征(receiver operating characteristic,ROC)曲线分析结果显示基于局部和全脑特征构建的急性脑卒中预后预测模型的曲线下面积(area under the curve,AUC)为0.900,敏感度和特异度分别为82.3%、89.1%,明显优于局部特征模型(AUC=0.706;Z=-3.248;P=0.001)和全脑特征模型(AUC=0.711;Z=-3.393;P<0.001)。结论联合局部和全脑的PWI特征可更为准确地预测急性脑卒中患者预后,为临床早期干预提供个性化指导。Objective:To investigate the predicting value of local and global brain radiomics of MR perfusion weighted imaging(PWI)in the outcome of acute stroke after endovascular.Materials and Methods:A total of 180 acute stroke patients with PWI images in our hospital were retrospectively enrolled.The ITK-SNAP software was used to segment the regions of interest of abnormal perfusion areas in Tmax.The SPM software was used to automatically segment the global brain in Tmax.The AK software was used to extract the local and global brain radiomics and reduce the dimensionality.The support vector machine classifier was used to construct the models for predicting the outcome of acute stroke,and further searching for the optimal prediction model.Results:After least absolute shrinkage and selection operator(LASSO)dimensionality reduction,there are 6 local features,5 global brain features,and 10 combined local and global brain features that are highly related to outcome.Receiver operating characteristic(ROC)analysis showed that area under the curve(AUC)of the outcome prediction model based on both local and global brain features was 0.900,with the sensitivity and specificity were 82.3%and 89.1%respectively,which is significantly better than the local features model(AUC=0.706;Z=−3.248;P=0.001)and the global brain features(AUC=0.711;Z=−3.393;P<0.001).Conclusions:Combining local and global PWI features can more accurately predict the outcome of acute stroke patients and provide personalized guidance for early clinical intervention.

关 键 词：卒中 预后 灌注加权成像 磁共振成像 影像组学 机器学习 

分 类 号：R445.2[医药卫生—影像医学与核医学] R743.3[医药卫生—诊断学]