应用经验模式分解去除近红外光的运动伪迹

Motion Artifact Removal for Near Infrared Spectroscopy by Empirical Mode Decomposition

针对近红外光谱(NIRS)技术对运动十分敏感和运动伪迹会严重影响后续分析处理结果这一问题,提出了一种利用经验模式分解去除运动伪迹的方法(EMD-MAR)。首先计算NIRS信号的移动标准偏差,根据其概率分布特性自动选取阈值和确定运动伪迹范围,然后利用经验模式分解(EMD)将信号分解成多个固有模式,对在运动伪迹范围具有明显异常的固有模式置0来消除运动伪迹,最后将处理后的固有模式加和得到校正后的信号。仿真与实测结果表明,EMD-MAR方法可有效检测和消除NIRS测量中常见的基线漂移、瞬时脉冲以及短暂低频振荡3种运动伪迹,并且极大提高了运动伪迹的自动检测程度,利用经验模式分解可以在去除运动伪迹的同时有效保留NIRS信号中的生理信息,为NIRS信号预处理提供了一种有效手段。

Near infrared spectroscopy （NIRS） is very sensitive to the movement of a body. The motion artifacts （MA） coupled to the optical measurement seriously affect the subsequent analysis results. Aiming at this problem, a method by empirical mode decomposition （EMD） to remove the motion artifacts （EMD-MAR） is presented. The moving standard deviation （MSD） of a NIRS signal is evaluated. From the characteristics of the probability distribution of the MSD, the thresholds of MA detection are determined and the ranges of the MAs are detected automatically. Then EMD is adopted to decompose the signal into intrinsic mode functions （IMF）. The values of the IMFs with obvious abnormal patterns within the detected range are set to zero to eliminate the MAs. All the IMFs treated are employed to reconstruct the corrected signal. The EMD-MAR is validated with simulated and real NIRS signals. The results show that it can effectively detect and eliminate the three kinds of MAs：baseline drift, transient impulse and transient oscillation, and a significant reduction of MAs and an increase in signal quality are achieved. The EMD-MAR method greatly improves the degree of automatic detection of motion artifacts to effectively keep the physiological information in a NIRS signal while the MAs are removed by the empirical mode decomposition.