Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (24): 276-282.DOI: 10.3778/j.issn.1002-8331.2006-0328

• Engineering and Applications • Previous Articles     Next Articles

Ensemble Method Classifies EEG from Stroke Patients

WANG Fang, ZHANG Xueying, HU Fengyun, LI Fenglian   

  1. 1.School of Information and Computer, Taiyuan University of Technology, Taiyuan 030024, China
    2.Department of Neurology, Shanxi Provincial People’s Hospital, Taiyuan 030012, China
  • Online:2021-12-15 Published:2021-12-13

集成分类器对脑卒中患者脑电的分类

王方,张雪英,胡风云,李凤莲   

  1. 1.太原理工大学 信息与计算机学院,太原 030024
    2.山西省人民医院 神经内科,太原 030012

Abstract:

The detection of Disorders of Consciousness(DoC) based on clinical experiments is time and manpower consuming and these experiments are intervalic. The Electroencephalogram(EEG) of stroke patients is collected to study the automatic classification of stroke patients with DoC and without DoC. As many as 9 kinds of quantitative EEG features are extracted to build brain network, and then the connectivity features of the brain network are imported into classifiers to classify DoC and no DoC in stroke patients. An Ensemble Of Support Vector Machine(EOSVM) is designed to solve the problem that classifiers always tend to the majority classes in the classification on imbalanced dataset. The experimental results show that the existing classifiers can classify the stroke patients with DoC and without DoC with an accuracy of about 70% and a sensitivity lower than 40%, however, the EOSVM classifier gives an accuracy of 96.79%, the sensitivity of 95.45% and specificity of 100%. These results show that the classifier EOSVM can improve the classification of imbalanced EEG dataset, and help the automatic detection of DoC in stroke patients.

Key words: stroke patients, ensemble classifier, quantitative EEG, brain network, disorders of consciousness, machine learning

摘要:

脑卒中患者意识障碍的检查和检测耗时耗力且非连续,采集脑卒中患者的脑电信号,以研究有意识障碍与无意识障碍的脑卒中患者的自动分类。对脑卒中患者的脑电图提取多达9种定量脑电特征,构建脑网络,将这些脑网络的连通性特征输入到分类器中,实现对脑卒中患者是否有意识障碍的分类。为解决非平衡数据集分类时严重偏向多数类的问题,设计集成支持向量机分类器。实验结果显示基于现有分类器的脑卒中意识障碍的分类正确率在70%左右,敏感度在40%以下;而基于集成支持向量机分类器的分类准确性可达96.79%,同时敏感度和特异性分别为95.45%和100%。实验结果表明集成支持向量机分类器对非平衡数据集的脑电分类准确率显著提升,并促进脑卒中患者意识障碍的自动识别。

关键词: 脑卒中, 集成分类器, 定量脑电图特征, 脑网络, 意识障碍, 机器学习