基于光电容积脉搏波描记法的无创血红蛋白浓度检测技术的研究

发布时间：2017-12-27 21:33

本文关键词：基于光电容积脉搏波描记法的无创血红蛋白浓度检测技术的研究　出处：《北京理工大学》2016年博士论文　论文类型：学位论文

【摘要】：无创血红蛋白浓度检测在手术、输血、献血、营养普查等方面具有重要的意义。传统血红蛋白浓度检测方法需要进行破皮取血,不仅给受测者造成了生理及心理上的负担,且无法连续、实时地监测血红蛋白浓度的变化情况。寻找无创、连续监测血红蛋白浓度变化趋势的理论、方法和技术实现是一个至今尚未解决的难题。本文旨在研究一种能够连续、无创、实时地监测血红蛋白浓度的新技术及其实现,主要工作与成果包括:介绍基于光电容积脉搏波描记法(PPG)的无创血红蛋白浓度检测技术的光学与生理学的相关基础知识。并根据修正的Lambert-Beer定律提出了基于多路PPG信号的无创血红蛋白浓度测量的数学模型。建立多路PPG信号采集系统平台,主要包括:传感器探头的设计与制作、模拟前端信号采集和处理模块设计与实现、上位机数据采集存储模块的实现。利用该系统平台完成了324例临床数据的采集,为后续PPG信号处理和血红蛋白浓度值预测提供了真实可靠的数据。研究PPG信号的信号处理算法,包括:滤除高频噪声、消除运动伪迹、基线漂移去除以及PPG信号特征信息提取算法等。降噪处理采用有限脉冲响应(FIR)滤波、无限脉冲响应(IIR)滤波以及滑动平均滤波算法,通过对含有高频噪声的PPG信号进行处理实验,表明三种方法都能有效滤除高频噪声。针对运动伪迹消除的问题,提出独立成分分析与自适应滤波相结合的方法和基于梳状滤波器滤波的方法,通过对仿真数据与真实PPG信号进行处理实验,证明此两种方法均能够有效滤除运动伪迹。针对基线漂移问题,采用中值滤波、小波分解以及希尔伯特-黄变换方法进行处理,通过对含有基线漂移的PPG信号进行处理分析,证明三种方法均能够在一定程度上滤除基线漂移成分。采集实验数据,并对样本数据进行回归建模研究。以PPG信号的特征信息作为输入,以血细胞分析仪检测到的血红蛋白浓度结果作为目标值进行回归建模。分别利用偏最小二乘(PLS)回归算法和主成分分析(PCA)与反向传播(BP)神经网络算法相结合的方法进行建模分析。通过利用PLS算法得到的回归模型对153例样本进行预测,得到均方根误差(RMSE)为1.56 g/dL,相关系数为0.56(p0.01)。通过利用PCA与BP神经网络算法相结合的方法得到的回归模型对62例样本进行预测,得到均方根误差为1.49 g/dL,相关系数为0.61(p0.01)。结果表明,所建立的回归模型能够较好地预测血红蛋白浓度。证明了通过本文设计的基于多路PPG信号的无创血红蛋白浓度检测系统能够实现血红蛋白浓度的无创检测。
[Abstract]:Noninvasive hemoglobin concentration detection is of great significance in the operation, blood transfusion, blood donation, nutrition survey and so on. Traditional hemoglobin concentration detection method needs skin and blood collection, which not only causes physiological and psychological burden to the subjects, but also cannot monitor the change of hemoglobin concentration continuously and in real time. It is a difficult problem to find the theory, method and technical realization of non-invasive and continuous monitoring of the change trend of hemoglobin concentration. This paper aims to study a kind of new technology can be continuous, non-invasive and real-time monitoring of hemoglobin concentration and its implementation, the main work and achievements include: introduction of optical plethysmography (PPG) based on the relevant basic knowledge of Optics and physiology of noninvasive hemoglobin concentration detection technology study. Based on the modified Lambert-Beer's law, a mathematical model of non invasive hemoglobin concentration measurement based on multiple PPG signals is proposed. A multi-channel PPG signal acquisition system platform is established, which mainly includes the design and fabrication of sensor probes, the design and implementation of analog front-end signal acquisition and processing module, and the realization of data acquisition and storage module of host computer. 324 cases of clinical data were collected by using the system platform, which provided real and reliable data for the follow-up PPG signal processing and the prediction of hemoglobin concentration. The signal processing algorithms of PPG signal are studied, including filtering high frequency noise, eliminating motion artifacts, removing baseline drift and extracting feature information of PPG signals. The finite impulse response (FIR) filtering, infinite impulse response (IIR) filtering and moving average filtering algorithm are adopted for noise reduction. Experiments on PPG signal with high frequency noise show that the three methods can effectively filter high frequency noise. Aiming at the problem of motion artifact elimination, a method combining independent component analysis with adaptive filtering and comb filter based filtering is proposed. Through the processing experiments of simulated data and real PPG signals, it is proved that all two methods can effectively remove motion artifacts. For the baseline drift problem, median filter, wavelet decomposition and Hilbert Huang transform are applied to deal with the baseline drift. By processing and analyzing the PPG signals with baseline drift, it is proved that the three methods can filter baseline drift components to a certain extent. The experimental data are collected, and the regression modeling of the sample data is studied. The characteristic information of PPG signal was used as input, and the result of hemoglobin concentration detected by blood cell analyzer was used as the target value for regression modeling. The method of partial least squares (PLS) regression and principal component analysis (PCA) and backpropagation (BP) neural network algorithm are used to model the modeling and analysis respectively. The regression model obtained by PLS algorithm is used to predict 153 samples, and the root mean square error (RMSE) is 1.56 g/dL and the correlation coefficient is 0.56 (P0.01). A regression model based on PCA and BP neural network algorithm is used to predict 62 samples. The root mean square error is 1.49 g/dL and the correlation coefficient is 0.61 (P0.01). The results show that the established regression model can predict the hemoglobin concentration better. It is proved that the noninvasive hemoglobin concentration detection system based on multiple PPG signals based on this paper can achieve noninvasive detection of hemoglobin concentration.
【学位授予单位】：北京理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：R446.1;TN911.7

【相似文献】