摘要
目的基于弥散加权成像(diffusion weighted imaging,DWI)的影像组学特征进行机器学习,构建预测急性脑卒中机械取栓后预后的模型。材料与方法回顾性分析在本院接受机械取栓治疗的急诊脑卒中患者,按随机数字表法分为训练集(n=252)和测试集(n=108)。采用A.K.软件提取DWI梗死区影像组学特征并应用最低绝对收缩和选择算子回归模型筛选最佳影像组学特征,基于所选特征通过支持向量机分类器建立急性脑卒中机械取栓后预后的预测模型,利用受试者操作特征(receiver operating characteristic,ROC)曲线评价模型的预测效能。结果每例患者的DWI图像提取1136个影像组学特征,降维后筛选出21个与预后高度相关的特征。ROC分析显示基于DWI模型预测训练集卒中患者机械取栓后预后的曲线下面积(area under curve,AUC)为0.956,敏感度和特异度分别为0.965、0.948,准确度达0.954;基于DWI模型预测测试集卒中患者机械取栓后预后的AUC为0.801,敏感度和特异度分别为0.818、0.816,准确度达0.828。结论基于治疗前DWI的影像组学特征和机器学习构建模型对急性脑卒中机械取栓后预后的预测具有较高的预测效能。
Objective:To construct a prediction model of outcome after mechanical thrombectomy in acute stroke by machine learning based on imaging omics characteristics of diffusion weighted imaging(DWI).Materials and Methods:Acute stroke patients in our hospital were retrospectively collected.These patients were divided into a training set(n=252)and a test set(n=108)according to random number table method.The imaging omics characteristics were extracted from lesions on DWI using A.K.software,and Least absolute shrinkage and selection operator regression model was used to screen the characteristics,and then,the selected characteristics were used to construct the prediction model by support vector machine classifier.Receiver operating characteristic(ROC)curve was used to evaluate the predictive efficacy of the model.Results:One thousand one hundred and thirty-six imaging omics characteristics of each patient were extracted from DWI,and 21 characteristics highly related to outcome after mechanical thrombectomy in acute stroke were screened after dimension reduction.ROC analysis showed that the area under curve(AUC)of DWI model in predicting outcome was 0.956 based on training set,the sensitivity and specificity were 0.965,0.948 respectively,and the accuracy was 0.954;the AUC of DWI model in predicting outcome was 0.801 based on test set,the sensitivity and specificity were 0.818,0.816 respectively,and the accuracy was 0.828.Conclusion:The imaging omics characteristics and machine learning model based on DWI before therapy can effectively predict outcome after mechanical thrombectomy in acute stroke.
作者
郭群
吴含
彭明洋
陈国中
殷信道
孙军
GUO Qun;WU Han;PENG Mingyang;CHEN Guozhong;YIN Xindao;SUN Jun(Department of Radiology,Nanjing First Hospital,Nanjing Medical University,Nanjing,21006)
出处
《磁共振成像》
CAS
CSCD
北大核心
2021年第10期32-35,48,共5页
Chinese Journal of Magnetic Resonance Imaging
基金
国家自然科学基金(编号:82001811)
江苏省自然科学基金(编号:BK20201118)。
关键词
卒中
弥散加权成像
影像组学
机器学习
预后
stroke
diffusion weighted imaging
imaging omics
machine learning
outcome
