基于单通道脑电信号的睡眠自动分期研究

Automatic sleep staging with a single electroencephalography channel

睡眠分期是睡眠评估的基础,在睡眠紊乱症的早期诊断和干预中起着重要的作用。本文利用集合经验模态分解对单通道脑电信号进行预处理,联合使用从分解得到的固有模态信号中提取的线性和非线性动力学等多元特性,构建了机器学习模型的输入特征空间,并最终训练出可行的睡眠自动分期模型。通过对111个健康受试者整夜睡眠数据的分期实验发现,使用本文提出的特征构建策略,能在多种经典的机器学习算法(反向传播神经网络、支持向量机、随机森林和极端梯度提升)中获得具有实用价值的睡眠自动分期模型。其中,基于极端梯度提升算法的模型在对睡眠状态进行4种分期和5种分期的任务中,准确率分别为81.0%和79.7%。

Sleep staging is the basis of scientific evaluation of sleep.It plays vital role in the early diagnosis and intervention of sleep disorders.The automatic sleep staging system was constructed by firstly using ensemble empirical mode decomposition(EEMD)to the electroencephalography(EEG)signals and then extracting both linear and nonlinear features from the decomposed intrinsic mode functions(IMFs).The features were fed into machine learning models based on five different algorithms.After applying the proposed method on sleep EEG data of 111 healthy subjects.It can be found that the proposed method can yield to automatic sleep staging systems with substantial performance on several machine learning models.Total accuracies of 81.0%and 79.7%were achieved by the model based on extreme gradient boosting algorithm,respectively in 4-label and 5-label staging.