稀疏恢复的梯度投影算法在PPG信号重建中的研究

摘要

Photoplethaysmography (PPG) is a noninvasive technique that reflects blood volume changes in the vessel and plays a vital role in the diagnosis of cardiovascular diseases.Nevertheless,it is extremely susceptive to motion artifact (MA) caused by the objects' intentional or unintentional movements and then affects the accuracy of the diagnosis results.To solve this problem,this paper employed the gradient projection for sparse recovery (GPSR) algorithm based on the compressed sensing (CS) theoretical framework.For the non-sparse signals,the CS theory mainly consist two steps:sparse and recovery.First,in order to ensure the significant information not be destroyed,the Haar wavelet base was used to find the best sparse field for sparse the signals.After that the MA was removed from the contaminated PPG signals while the sparse signals were recovered.During the experiment,three types of noise were measured,and then GPSR algorithm was used to process the noisy signals,results showed that the GPSR algorithm significantly reduced the movement interference in the signals.In order to make the conclusions more convincing,other 50 healthy adults with vertical movement of MA were measured,the heart rate and mean square error (MSE) of the original and reconstruct PPG signals were calculated.Comparing the obtained heart rates with the noiseless PPG signals' heart rates through the Bland-Altman analysis method,it was shown that the range of the differences between the noisy and noiseless PPG signals were about ± 23 beat/ min,and the reconstructed PPG signals were about ±2.7 beat/min.The Box Plot chart indicated that the MSE of the reconstructed signals was reduced about 50％ compared with that of the noisy signals.%光电容积描记(PPG)法是一种无创测量人体脉搏信号的方法,其波形反映血管中的血液容积变化,在诊断心血管疾病时具有重要作用.然而,这种方法容易因病人有意或无意运动产生运动伪影(MA)而影响诊断结果的准确性.为解决这一问题,基于压缩感知理论框架,采用稀疏恢复的梯度投影(GPSR)算法,对PPG信号进行滤噪和恢复.对于非稀疏信号,压缩感知包括稀疏和恢复两个步骤:首先为保证信号的重要信息不被破坏,利用哈尔小波基找到最佳稀疏域,对信号进行稀疏变换;然后采用GPSR算法,在恢复信号的同时去除MA噪声.实验测量3种受MA影响的带噪PPG信号,并用GPSR算法处理分析.另测50名健康成年人在垂直运动时产生的带噪PPG信号,计算带噪与重建PPG信号的心率和均方误差(MSE).将得到的两种心率分别与无噪声PPG信号心率通过Bland-Ahman方法做对比分析,结果显示,带噪与无噪声PPG信号的差异范围在±23 beat/min,而重建PPG信号在±2.7 beat/min.由均方误差的Box Plot图可知,重建信号的MSE比带噪信号降低约50％.这些结果均表明,GPSR算法能明显降低信号中的运动干扰.