吉林大学学报(医学版) ›› 2023, Vol. 49 ›› Issue (6): 1635-1641.doi: 10.13481/j.1671-587X.20230631

• 影像学 • 上一篇    下一篇

机器学习结合磁共振弥散加权成像识别发病4.5 h内急性脑卒中患者

徐灿飞1,2,黄艳3,张开治1,3()   

  1. 1.吉林大学中日联谊医院神经外科,吉林 长春 130033
    2.浙江大学医学院附属邵逸夫医院病理科,浙江 杭州 310020
    3.吉林大学中日联谊医院医疗保险管理部,吉林 长春 130033
  • 收稿日期:2023-03-29 出版日期:2023-11-28 发布日期:2023-12-22
  • 通讯作者: 张开治 E-mail:zhangkz@jlu.edu.cn
  • 作者简介:徐灿飞(1994-),男,浙江省杭州市人,医学硕士,主要从事神经外科相关医学影像和人工智能等方面的研究。
  • 基金资助:
    吉林省卫健委卫生科研人才专项项目(2020SCZ25)

Acute ischemic stroke patients with onset time within 4.5 h identified by machine learning combined with magnetic resonance diffusion weighted imaging

Canfei XU1,2,Yan HUANG3,Kaizhi ZHANG1,3()   

  1. 1.Department of Neurosurgery,China-Japan Union Hospital,Jilin University,Changchun 130033,China
    2.Department of Pathology,Sir Run Run Shaw Hospital,School of Medical Sciences,Zhejiang University,Hangzhou 310020,China
    3.Department of Medical Insurance Administration,China-Japan Union Hospital,Jilin University,Changchun 130033,China
  • Received:2023-03-29 Online:2023-11-28 Published:2023-12-22
  • Contact: Kaizhi ZHANG E-mail:zhangkz@jlu.edu.cn

摘要:

目的 探讨机器学习方法结合磁共振弥散加权成像(DWI)识别4.5 h内急性缺血性脑卒中(AIS)患者的效果,为辅助评估AIS患者发病时间提供参考。 方法 选择DWI影像资料完整的AIS患者227例,根据发病至DWI检查时间将患者分为发病时间≤ 4.5 h组(n=70)和发病时间>4.5 h组(n=157)。227例患者采用完全随机法按照7∶3的比例划分为训练集(n=158)和测试集(n=69)。采用ITK-SNAP标注软件于DWI图像上划分感兴趣区域(ROI),采用Python软件Pyradiomics包由ROI图像中提取图像特征,Spearman相关性检验评估各特征的相关性并去除一致性过高的冗余特征,结合10倍交叉验证的最小绝对收缩和选择算子(LASSO)回归模型筛选可用于识别发病4.5 h内AIS患者的图像特征,采用支持向量机(SVM)、K近邻 (KNN)、决策树、极限树、随机森林、XGBoost和LightGBM共 7种机器学习算法训练模型,采用受试者工作特征(ROC)曲线和曲线下面积(AUC)评估模型性能。 结果 由ROI图像中共提取107个图像特征,包括18个一阶特征、14个形态特征和75个纹理特征,经筛选最终获取22个可用于识别发病4.5 h内AIS患者的图像特征,包括4个一阶特征、6个形态特征和12个纹理特征。XGBoost模型识别发病4.5 h内AIS患者效果最佳,AUC为0.817,准确度、灵敏度和特异度分别为0.739、0.733和0.814。 结论 基于DWI影像的XGBoost模型可有效识别发病4.5 h内的AIS患者,对辅助评估AIS患者发病时间有一定的临床意义。

关键词: 急性缺血性脑卒中, 机器学习, 磁共振弥散加权成像, 发病时间, XGBoost模型

Abstract:

Objective To discuss the effectiveness of machine learning method combined with magnetic resonance diffusion weighted imaging (DWI) for recognition of the acute ischemic stroke (AIS) patient with onset time within 4.5 h,and to provide the reference for assisted assessment of onset time of AIS patients. Methods A total of 227 AIS patients with complete DWI imaging data were divided into onset time≤ 4.5 h group (n=70) and onset time >4.5 h group (n=157) based on their time from onset to DWI examination. The patients were randomly divided into training set (n=158) and test set (n=69) at a ratio of 7∶3. The regions of interest (ROI) were designated on the DWI images by ITK-SNAP annotation software, and the image features were extracted from the ROI images by the Pyradiomics package.The redundant features with high consistency were removed after evaluating the correlation of each feature by Spearman correlation test. Least absolute shrinkage and selection operator (LASSO) regression model with 10-fold cross-validation was used to recognize the image features of the AIS patients with onset time within 4.5 h. Seven machine learning algorithms, including Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Decision Tree, Extra Trees, Random Forest, XGBoost, and LightGBM, were used to train the models. The performance of the models was evaluated by receiver operating characteristic curve (ROC) and the area under the curve (AUC). Results A total of 107 image features, including 18 first-order features, 14 shape features, and 75 texture features, were extracted from the ROI images, among which 22 features were finally selected for recognition of the AIS patients with onset time within 4.5 h, including 4 first-order features, 6 shape features, and 12 texture features. The XGBoost model yielded the best results in recognizing the AIS patients with onset time within 4.5 h, and the AUC was 0.817, and the accuracy, sensitivity, and specificity were 0.739, 0.733, and 0.814,respectively. Conclusion The XGBoost model based on DWI imaging can effectively recognize the AIS patients with onset time within 4.5 h, which has certain clinical significance in assisting the assessment of onset time of the AIS patients.

Key words: Acute ischemic stroke, Machine learning, Magnetic resonance diffusion weighted imaging, Onset time, XGBoost model

中图分类号: 

  • R743.3