经验模态分解的单通道呼吸信号自动睡眠分期

Empirical Modal Decomposition of Single-Channel Respiratory Signals for Automatic Sleep Staging

摘要: 睡眠是人体基本的生理需求，可以保证机体的生长发育、为机体储蓄能量、维持机体免疫等。对睡眠质量的准确评估是认识睡眠障碍并采取有效干预措施的关键。如果用经验丰富的睡眠专家进行人工睡眠分期是比较耗时并且主观的。目前，研究人员提出了许多准确、有效、有针对性的睡眠分期方法。比如，基于深度学习以及经验模态分解算法的单通道电脑信号自动睡眠分期方法，它被成功地用于呼吸信号(RESP)的睡眠分期，该方法为呼吸信号分解和睡眠阶段自动识别提供了新途径。本文采用的呼吸信号数据集来自SHHS，它是一个中心队列研究，用来确定睡眠与呼吸障碍的心血管和其他病症的数据库。首先，我们对SHHS数据库中的单通道呼吸信号进行了分析，以便更好地了解人类睡眠情况。其次，利用经验模态分解算法(EMD)对预处理后的呼吸信号进行分解，从原始呼吸信号和分解出的6个简单信号中提取时域、非线性动力学、统计学等方面的9个特征。最后，使用长短期记忆网络(LSTM)构建分类模型，将提取的呼吸信号特征进行分类识别，实现自动睡眠分期。实验结果表明，在4类和5类睡眠分期任务中，SHHS数据库的呼吸信号自动睡眠分期准确率分别为89.22%和88.43%。实验结果表明，本文提出的自动睡眠分期模型具有较高的分类精度和效率，具有较强的适用性和稳定性。

Abstract: Sleep is a basic physiological need of the body to ensure growth and development, save energy for the body, and maintain immunity of the body. Accurate assessment of sleep quality is the key to recognizing sleep disorders and taking effective interventions. Manual sleep staging is time con-suming and subjective when performed by experienced sleep specialists. Currently, researchers have proposed a number of accurate, effective, and targeted sleep staging methods. For example, a single-channel computer signal automatic sleep staging method based on deep learning and empir-ical modal decomposition algorithms has been successfully used for respiratory signal (RESP) sleep staging, which provides a new way to decompose respiratory signals and identify sleep stages au-tomatically. The data set used in this paper is from SHHS, which is a central cohort study to identify sleep and breathing disorders in a database of cardiovascular and other conditions. First, we ana-lyzed the single-channel respiratory signals from the SHHS database to better understand human sleep. Second, the pre-processed respiratory signals were decomposed using an empirical modal decomposition algorithm (EMD) to extract nine features in the time domain, nonlinear dynamics, and statistics from the original respiratory signals and the six simple signals that were decomposed. Finally, a classification model was constructed using a long short-term memory network (LSTM) to classify and identify the extracted respiratory signal features for automatic sleep staging. The ex-perimental results show that the accuracy of automatic sleep staging of respiratory signals from SHHS database is 89.22% and 88.43% in 4 and 5 categories of sleep staging tasks, respectively. The experimental results show that the automatic sleep staging model proposed in this paper has high classification accuracy and efficiency, and has strong applicability and stability.