睡眠呼吸暂停综合征患者的睡眠分期算法研究

针对可穿戴睡眠监测缺乏有效的自动睡眠分期和睡眠质量评价方法这一问题,提出一种适用于睡眠呼吸暂停综合征患者的自动睡眠分期方法。通过心电图R-R间期序列,分别得到心率变异性、呼吸幅度变异性和呼吸率变异性信号。以此为基础,提取时域、频域及非线性特征共55个。利用门控循环单元网络,分别构建清醒-睡眠二分类、清醒-快速眼动-非快速眼动睡眠三分类、清醒-快速眼动-浅睡-慢波睡眠四分类、清醒-快速眼动-非快速眼动Ⅰ-Ⅱ-Ⅲ期五分类等共4个不同分类粒度的睡眠分期模型;采用损失函数类别加权方法,有效降低数据非平衡对分期结果的影响。验证数据来自SHRS数据库的274例患者。借助准确率、Cohen's Kappa系数和睡眠结构指标对该睡眠分期方法进行性能评价。结果表明4个分类器的准确率分别为85.06%、75.44%、63.80%、62.13%,Cohen's Kappa系数达到了0.54、0.49、0.41、0.41,睡眠结构分析评估与临床结果之间的差异无统计学意义。所提出的方法基本满足睡眠质量评估的需求,适用于可穿戴睡眠监测应用。

A Study on Sleep Staging Algorithm for Patients with Sleep Apnea Syndrome

Aiming at the lack of effective automatic sleep staging method for wearable sleep monitoring, an automatic sleep staging algorithm suitable for patients with sleep apnea syndrome was proposed in this paper. Through the R-R interval sequence of an ECG signal, the signals of heart rate variability (HRV), respiratory amplitude variability (RAV) and respiratory rate variability (RRV) were obtained. On this basis, 55 features of time, frequency, and nonlinear domains were extracted. Four sleep staging models with different classification granularity were constructed by using gated circulation unit network: W-S two classification, W-REM-NREM three classification, W-REM-LS-SWS four classification, W-REM-N1-N2-N3 five classification. A categorical weighting method for the loss function was used to effectively reduce the impact of data imbalance on the sleep staging results. The accuracy, Cohen's kappa coefficient and sleep structure indexes were used to evaluate the performance of the sleep staging algorithm. The results showed that the values of accuracy and kappa of the four classifiers were 85.06%, 75.44%, 63.80%, 62.13%, and 0.54, 0.49, 0.41, 0.41, respectively, and the sleep structure analyzing showed there was no statistically significant difference from clinical results. The method proposed displayed the ability of meeting the needs of sleep quality evaluation and is expected to be used for wearable sleep monitoring applications.