文献类型：期刊文献

中文题名：基于贝叶斯最小风险的癫痫脑电自动检测算法

英文题名：Automatic detection of epileptic EEG based on minimum Bayesian risk and rotation forest

作者：卫作臣[1];邹俊忠[1];张见[1];陈兰岚[1]

机构：[1]华东理工大学信息科学与工程学院自动化系

年份：2019

卷号：36

期号：12

起止页码：3729

中文期刊名：计算机应用研究

外文期刊名：Application Research of Computers

收录：CSTPCD;;北大核心:【北大核心2017】；CSCD:【CSCD_E2019_2020】；

基金：国家自然科学基金资助项目(61201124);中央高校基本业务资金资助项目(222201817006)

语种：中文

中文关键词：癫痫;时域特征;随机映射;旋转森林;代价敏感;贝叶斯最小风险

外文关键词：epilepsy;time-domain feature;random projection;rotation forest;cost sensitive;minimum Bayesian risk

摘要：提出一种新的不平衡分类算法,基于增减序列合并周期分割算法提取时域特征,引入随机映射优化了旋转森林的计算效率,进而计算基于海林格距离的贝叶斯最小风险来给出测试样本标签。该算法在1 s片段上得到了90. 66%灵敏性,92. 52%特异性,F2分数为0. 905 5,并且检出了98. 56%的癫痫发作,检测延迟为1. 32 s,在不平衡的癫痫脑电数据集上表现出了良好的性能,对于癫痫辅助诊断有着极大的临床意义。
This paper proposed a novel automatic epileptic EEG detection approach. This study calculated the time domain features based on the merger of increasing and decreasing sequence( MIDS),and employed random projection to improve the complexity of rotation forest,as well as predicted the sample label using minimum Bayesian risk based on the Hellinger distance. This approach yielded 90. 66% sensitivity,92. 52% specificity and F2-score of 0. 905 5 in EEG segment classification task. Moreover,the proposed approach achieved 98. 56% sensitivity of seizures with the average latency 1. 32 s. The proposed method shows good performance on the epileptic EEG imbalanced data,and has great clinical significance for the auxiliary diagnosis of epilepsy.