基于单通道ECG信号与INFO-ABCLogitBoost模型的睡眠分期

Sleep staging based on single-channel ECG signal and INFO-ABCLogitBoost model

为了减少对传统多导睡眠图(PSG)系统的依赖,基于单通道心电图(ECG)信号,设计了一种简单高效的睡眠分析算法.采用最大重叠离散小波变换(MODWT)对原始信号进行多分辨分析,再进一步提取峰值信息;根据峰值位置的一阶偏差,提取多维度的心率变异性(HRV)特征.为了进一步筛选与不同睡眠阶段具有强关联性的HRV特征,提出基于ReliefF算法与Gini指数的特征提取方法.在此基础上,采用INFO-ABCLogitBoost方法挖掘HRV与不同睡眠阶段之间的关联性,从而实现睡眠阶段的精细分类.在实际公开数据集上的实验结果表明,所提出的模型在睡眠分期任务中,总体精度为83.67%,准确率为82.59%,Kappa系数为77.94%,F1-Score为82.97%.相比于睡眠分期任务中的常规模型,所提方法展现出更加高效便捷的睡眠质量评估性能,有助于实现家庭或移动医疗场景下的睡眠监测.

A simple and efficient sleep analysis algorithm was designed based on single-channel electrocardiogram(ECG)signals,in order to reduce the dependence on polysomnography(PSG)system.First,maximum overlap discrete wavelet transform(MODWT)was used to perform multi-resolution analysis on the original signal,then to furture extract peak information.Then,the multi-dimensional heart rate variability(HRV)features were extracted based on the first-order deviation of the peak position.To further screen the HRV features with a strong correlation with different sleep stages,a feature extraction method was proposed based on the ReliefF algorithm and Gini index.On this basis,the INFO-ABCLogitBoost method was used to mine the correlation between HRV and different sleep stages,thereby achieving a fine classification of sleep stages.Experimental results on actual public data sets showed that the proposed model had an overall accuracy of 83.67%,an accuracy rate of 82.59%,a Kappa coefficient of 77.94%,and an F1-Score value of 82.97%in the sleep staging task.Compared with conventional models in sleep staging tasks,the proposed method shows more efficient and convenient sleep quality assessment performance,which helps realize sleep monitoring in home or mobile medical scenarios.