基于序列连通度的睡眠分期算法研究

Sleep Staging from the Visibility Graph Algorithm of Series

准确的睡眠分期有利于帮助人们改善睡眠质量.本文提出了一种基于序列连通度分析的特征参数提取算法,提取了连通度分布斜率,连通距离均值,平均连通距离均值以及改进的加权连通度均值等特征参数,采用最小二乘支持向量机对其进行训练和学习,建立了睡眠脑电的数学模型.结果表明,相对于目前已有的序列加权连通度算法,本文算法对于不同睡眠状态的分期正确率提高了约5.72%,特别是对于浅睡眠状态的分类正确率提高约9.65%.

Monitoring the sleep quality accurately can play an effective supporting role in helping people improve the quality of sleep. In the present study,a novel feature extraction algorithm is proposed based on the natural visibility graph and horizontal visibility graph methods. The slope of visibility degree distribution,the mean of visibility distance,the mean of averaged visibility distance and the mean of improved weighted visibility graph were extracted,and trained by the least square-support vector machines（ LS-SVM） classifier. The mathematical model between electroencephalogram（ EEG） and sleep state was established and verified by different samples. The results demonstrated that the classification accuracy of different states improved about 5. 72% compared to the existing weighted visibility graph,the classification accuracy of shallowsleep states improved about 9. 65%.