摘要
目的分析基于恶性大脑中动脉梗死(mMCAI)相关的超急性期MRI影像组学特征的机器学习在mMCAI预测方面的价值。方法对243例大脑中动脉(MCA)-M1段闭塞的超急性期脑梗死患者行头颅MRI检查。采用ITK-SNAP软件对超急性期MRI图像上的DWI高信号梗死区和Flair对应高信号区进行分割,采用AK软件进行影像组学特征提取和降维,最终使用最小绝对收缩与选择算子算法(LASSO)确定mMCAI相关的超急性期影像组学特征,通过支持向量机分类器评估其在mMCAI预测中的价值。结果每例患者提取792个影像组学特征,降维后筛选出10个与mMCAI预测最相关的特征。ROC分析显示,该模型预测训练集患者mMCAI的曲线下面积(AUC)为0.986,灵敏度和特异度分别为0.988、0.928;预测测试集患者mMCAI的AUC为0.972,灵敏度和特异度分别为0.949、0.889。结论基于MRI的影像组学和机器学习分析是预测脑梗死患者恶性进展的重要工具,对早期准确识别mMCAI高风险患者具有较高的效能。
Objective To analyze the value of machine learning based on MRI radiomic features in predicting the malignant middle cerebral artery infarction(mMCAI)in the hyperacute phase.Methods A total of 243 infarction patients with hyperacute MCA⁃M1 segment occlusion were examined by cranial MRI.The ITK⁃SNAP software was used to segment the high signal areas of DWI and the corresponding high signal areas of Flair.The AK software was used to extract the radiomic features and reduce the dimensionality.The LASSO regression analysis was used to determine the radiomic features related to mMCAI and support vector machine classifier was used to evaluate its value in mMCAI prediction.Results Seven hundred and ninety⁃two radiomics features of each patient were extracted and 10 features highly related to mMCAI were screened after dimension reduction.ROC analysis showed that the area under curve(AUC)of the prediction model based on the training set was 0.986,the sensitivity and specificity were 0.988 and 0.928 respectively;the AUC of the prediction model based on the test set was 0.972,the sensitivity and specificity were 0.949 and 0.889 respectively.Conclusion The analysis based on MRI radiomics and machine learning are the important tools for predicting mMCAI,and have high efficiency in early accurate identification of mMCAI.
作者
刘景明
彭明洋
王同兴
陈国中
谢光辉
马跃虎
LIU Jingming;PENG Mingyang;WANG Tongxing(Department of Radiology,Xuzhou Tumor Hospital,Xuzhou 221005,China)
出处
《临床神经病学杂志》
CAS
2023年第4期241-246,共6页
Journal of Clinical Neurology
基金
国家自然科学基金(82001811)。
关键词
脑梗死
恶性
MRI
影像组学
机器学习
cerebral infarction
malignant
MRI
radiomics
machine learning
