摘要
针对目前警觉性监测指标单一且不精确等问题,将心率变异性结合机器学习对警觉度水平进行监测。48名受试者连续完成5组精神运动警戒任务(PVT),同时进行心电测量。将PVT任务的反应时分为5级,使用多分类支持向量机方法,对警觉性分级进行预测。在行为结果上,表现为PVT测试反应时相关指标的延长;在生理指标上,反映为心率变异性高频指标(HF-HRV)以及极低频指标(VLF-HRV)的增加。VLF-HRV与反应时的平均数以及反应时的中位数呈现显著的正相关,与反应时倒数的平均值呈现出显著的负相关。HF-HRV也与反应时的平均数以及反应时的中位数呈现显著的正相关。采用多分类支持向量机手段对受试者的警觉度进行预测,结果表明,单独使用2种心电指标对警觉度水平进行预测的平均准确率为77.81%,ROC曲线下的平均面积为0.87,平均灵敏度为0.763,平均特异度为0.792。研究表明:心率变异性是反映警觉度波动变化的敏感指标,可用于开展警觉度的预测。
Till now,there are few indexes for precise prediction of vigilance fluctuation.In this study,heart rate variability and machine learning were combined together to detect vigilance level.48 subjects were recruited to finish a 5-block psychomotor vigilance task(PVT).Meanwhile,electrocardiography(ECG)data was collected during the experiment.PVT reaction time was divided into five levels and multi-class support vector machine was used to predict the vigilancelevel by heart rate variability.PVT reaction time became longer during the task,which was accompanied by increase of HF-HRV and VLF-HRV.VLF-HRV was negatively correlated with mean of reciprocal reaction time and positively correlated with median of reaction time.While HF-HRV was positively correlated with both mean and median of reaction time.In the end,multi-class support vector machine was used to predict the vigilance level by HF-HRV and VLFHRV.The results showed that the mean accuracy was 77.81%;the area under ROC curve was 0.811;the mean sensitivity was 0.763;and the mean specificity was 0.792.It was concluded that heart rate variability was sensitive to the fluctuation of vigilance level and could be utilized in prediction of vigilance.
作者
周维逸
周仁来
ZHOU Weiyi;ZHOU Renlai(Department of Psychology,Nanjing University,Nanjing 210023,China)
出处
《载人航天》
CSCD
北大核心
2022年第6期779-784,共6页
Manned Spaceflight
基金
载人航天工程航天医学实验领域项目(HYZHXM03008)。
关键词
警觉度
心率变异性
机器学习
生理指标
vigilance
heart rate variability
machine learning
physiological index
