基于有限穿越水平可视图的短时睡眠心率变异性研究

心率数据是最易于获取的人体生理数据之一,基于心率变异性的睡眠分析是近年来各种用于日常健康管理的可穿戴设备功能的一个重要发展方向,需要不断探索可以应用于标准睡眠分期时间窗(约30 s)的各类短时特征参数.利用近期报道的有限穿越水平可视图,并进一步提出一种加权有限穿越水平可视图,将不同睡眠状态下的短时心率变异序列映射为网络,进而提取平均集聚系数、特征路径长度、集聚系数熵、路径分布熵、加权集聚系数熵和加权路径分布熵等网络特征参数进行统计分析.结果表明,各网络参数值在醒觉、浅睡期、深睡期和快速眼动期的幅度水平具有显著差异,体现了所述方法在基于短时心率变异数据的睡眠分期中的有效性.同时,进一步研究了健康年轻人和中老年人在不同睡眠状态下的网络参数值,发现两者虽然存在整体的水平差异,但是在不同睡眠状态间的变化仍具有相同的趋势,反映出相对于正常的年龄老化,睡眠调制对心脏动力学系统具有更显著的影响,也说明所述方法可作为基于心率变异性的睡眠研究的一种新的辅助工具.

Research of short-term heart rate variability during sleep based on limited penetrable horizontal visibility graph

Heart rate is one of the most easily accessed human physiological data. In recent years, the analysis of sleep function based on heart rate variability has become a new popular feature of wearable devices used for daily health management. Consequently, it is needed to explore various types of short-term characteristic parameters which can be applied to the heartbeat interval time series within the standard sleep staging time window (about 30 s). Utilizing the recently reported limited penetrable horizontal visibility graph (LPHVG) algorithm, together with a weighted limited penetrable horizontal visibility graph (WLPHVG) algorithm proposed in this paper, the short-term heartbeat interval time series in different sleep stages are mapped to networks respectively. Then, 6 characteristic parameters, including the average clustering coefficient C, the characteristic path length L, the clustering coefficient entropy E_c, the distance distribution entropy E_d, the weighted clustering coefficient entropy E_Cw and the weight distribution entropy E_w are calculated and analyzed. The results show that the values of these characteristic parameters are significantly different in the states of wakefulness, light sleep, deep sleep and rapid eye movement, especially in the case of the limited penetrable distance L_p=1, indicating the effectiveness of LPHVG and WLPHVG algorithm in sleep staging based on short-term heartbeat interval time series. In addition, a preliminary comparison between proposed algorithm and the basic visibility graph (VG) algorithm shows that in this case, the LPHVG and WLPHVG algorithm are superior to the basic VG algorithm both in performance and in calculation speed. Meanwhile, based on the LPHVG and WLPHVG algorithm, the values of network parameters (the clustering coefficient entropy E_c and the weighted clustering coefficient entropy E_Cw) are calculated from heartbeat interval time series of healthy young and elder subjects in different sleep stages, to further study the aging effect on and sleep regulation over cardiac dynamics. It is found that despite an overall level difference between the values of E_c and E_Cw in young and elder groups, the stratification patterns across different sleep stages almost do not break down with advanced age, suggesting that the effect of sleep regulation on cardiac dynamics is significantly stronger than the effect of healthy aging. In addition, compared with the clustering coefficient entropy E_c based on LPHVG algorithm, the weighted clustering coefficient entropy E_Cw based on WLPHVG algorithm shows higher sensitivity to discriminating subtle differences in cardiac dynamics among different sleep states. Overall, it is shown that with the simple mapping criteria and low computational complexity, the proposed method could be used as a new auxiliary tool for sleep studies based on heart rate variability, and the corresponding network parameters could be used in wearable device as new auxiliary parameters for sleep staging.