基于心率和呼吸特征结合的睡眠分期研究

目的研究通过心率和呼吸信号进行睡眠分期,探讨将心率和呼吸信号结合进行睡眠分期研究的有效性。方法基于ISURE-sleep睡眠数据库中9名健康受试者ECG和呼吸波数据,采用时域、频域和非线性分析方法,计算并筛选出HRV、呼吸波和心肺耦合特征共计34个指标。基于SVM、随机森林、XGBoosting和BP神经网络4种分类器构建模型,采用独立被试和非独立被试两种方案对觉醒期、快速眼动期和非快速眼动期进行分类测试。结果在4种分类模型下,综合建模分期方法结果均优于基于HRV或呼吸单一信号建模的睡眠分期方法,综合建模Kappa系数有显著提升。基于XGBoosting的模型在4种方法中表现最优,取得了独立被试测试73.3%的平均准确率(Kappa=0.42)和非独立被试测试准确率88.7%(Kappa=0.75)。结论加入呼吸特征和心肺耦合特征可以作为辅助指标,提升目前常用的基于HRV的睡眠分期模型的性能。心率和呼吸特征结合的睡眠分期方法在便携式睡眠监测等领域具有实际应用价值。

Sleep Staging Based on Combination of Heart Rate and Respiratory Characteristics

Objective To study the feasibility of sleep staging by combining heart rate and respiratory signal.Methods Based on the ECG and respiratory wave data of 9healthy subjects in ISURE-sleep sleep dataset,34HRV,respiratory wave and cardiopulmonary coupling characteristics were calculated and screened by using time-domain,frequency-domain and nonlinear analysis methods.Based on SVM,random forest,xgboosting and BP neural network,the models were constructed.Independent subjects and non independent subjects plans were used to test the wake,rapid eye movement and non rapid eye movement period.Results Under the four classification models,the comprehensive modeling method was superior to the HRV based or breathing single signal based sleep staging method,especially the improvement of Kappacoefficient.The model based on xgboosting had the best performance among the four methods,and got an average accuracy of 73.3%for independent subjects plan(Kappa=0.42)and 88.7%for non independent subjects plan(Kappa=0.75).Conclusion Respiratory characteristics and cardiopulmonary coupling characteristics can be used as auxiliary indexes to improve the performance of HRV based sleep staging model.The sleep staging method combined with heart rate and respiratory characteristics has practical application value in portable sleep monitoring and other fields.