脉搏信号获取与分析的研究

发布时间：2018-05-08 21:51

本文选题：脉搏信号采集系统 + 脉搏信号预处理　；参考：《哈尔滨工业大学》2014年博士论文

【摘要】：脉搏信号是一种重要的人体生理信号,包含了丰富的循环系统信息。由于脉搏的搏动在体表就能触摸到,因而脉搏信号的获取具有无痛苦、采集简单、成本低廉等优势。脉搏信号的分析自古就备受医学界青睐,在古代中医、古代印度医学、古埃及医学中脉诊都占有十分重要的地位。在利用脉诊诊病时医师将手指放置于患者桡动脉的寸、关、尺三个穴位,通过指腹的触感来感受不同取脉压力下脉搏的搏动从而分析患者的身体状况。然而,脉诊的技巧需要长期的学习和实践才能掌握,而且这种诊法的主观性较强,存在不同医师诊断结果不一致的问题。为了克服这些问题,近年来越来越多的学者利用传感器技术获取脉搏信号并利用计算机技术来分析病人的健康状况,使脉诊技术向客观化的方向发展。在脉搏信号的获取与分析中主要包括脉搏信号的获取、预处理、特征提取与模式分类等内容,本文从这四方面出发对脉搏信号的获取与分析展开了系统的研究。脉搏信号的获取是脉搏信号分析研究的基础。近年来不同的研究机构研发了各种各样的脉搏信号获取系统,然而现有的采集设备仍存在一些不足以待改进,首先现有的脉搏信号采集系统大多没有定位辅助系统,将传感器放置于患者手腕的过程仍然很大程度依赖采集人员的经验,使得位置的选择较为主观和费时。其次现有的系统不能自动控制采样压力,使得压力的设置较为主观和费时。另外现有设备在测量脉宽和寸、关、尺三个部位同时采集等方面也存在一些欠缺。针对这些问题,本文设计了新的脉搏信号获取系统。通过该系统可以快速的找到合理的采样位置并根据预设的压力自动调节探头高度,减少了采集过程中的主观因素并加快了采样速度,提高了采样的客观性。本文的系统还具有测量脉宽和寸、关、尺三路同时采集的功能。脉搏信号的预处理旨在去除耦合在脉搏信号中的干扰,提升脉搏信号的质量从而提高后续的特征提取以及模式分类的准确性。现有的预处理过程通常关注去除脉搏信号中耦合的噪声和基线漂移等干扰,然而有一些干扰由于发生了信息损失或者与脉搏信号的频带重叠使得这类干扰的处理去除难度较大,如饱和、伪迹这两种干扰很难通过现有的预处理方法得到处理。这使得预处理之后脉搏数据集中包含异常样本,从而影响了特征提取和分类的准确性。针对这个问题本文扩展了原预处理框架,增加了饱和检测模块和伪迹检测模块,并提出了基于差分的饱和检测算法和基于复杂网络连通性的伪迹检测算法,可以有效的检测饱和与伪迹这两种常见干扰。脉搏信号的特征提取是脉搏信号分析的另一个重要课题。目前常用的脉搏信号特征大体上可以分为两类:一类是以分析脉搏周期的特点为出发点的基于单个脉搏周期的特征提取方法;一类是以分析脉搏信号整体性质为出发点的基于整个脉搏信号的脉搏特征提取方法。脉搏信号是一个准周期信号,各个周期之间存在一定的差异,我们认为这种差异具有一定的诊断意义尤其是对一些与节律相关的疾病的诊断。然而这两种方法都没有注重对脉搏信号周期间差异的刻画,本文从刻画脉搏信号周期间差异出发提出了三种对周期间差异敏感的特征。实验结果表明周期间的差异对于疾病诊断具有重要意义,尤其是在一些对周期间差异较敏感的疾病的诊断。由于本文同时考虑了脉搏信号周期间和周期内的差异,在分类实验中本文所提取的特征获得了相对于其他特征更好的识别效果。在脉搏信号的分类研究中本文比较了压力脉搏信号与另外两种较为常见的脉搏信号(光电脉搏信号和超声脉搏信号)的分类性能,并提出基于组合核模型的脉搏信号融合分类算法。通过分析三种不同的脉搏信号的的采集原理、物理意义及其相互联系,本文讨论了不同类型脉搏信号各自的敏感特征,我们认为不同类型的脉搏信号对于不同疾病的诊断效果有一定的差别,如果某种疾病所引起的病变与某种类型脉搏信号的敏感特征相关,则在这种疾病的诊断中该类型的脉搏信号具有一定的诊断精度优势。实验结果表明,超声脉搏信号在糖尿病的诊断中取得了相对其他脉搏信号更好的诊断效果,而压力信号在动脉硬化的诊断上取得了更好的诊断结果。利用信号之间的互补性,融合使用多种脉搏信号我们可以进一步获得更多有效地诊断信息。为此我们提出了基于组合核模型的多种脉搏信号的融合分类算法,并在实验中取得了比使用单一脉搏信号更高的诊断精度。
[Abstract]:Pulse signal is an important physiological signal of human body. It contains abundant information of circulatory system. Because pulse pulsation can be touched on the body surface, pulse signal has the advantages of painless, simple collection, low cost and so on. Pulse signal analysis has attracted much attention from medical circles since ancient times, in ancient Chinese medicine, ancient India medicine, Pulse diagnosis in ancient Egyptian medicine occupies a very important position. In the use of pulse diagnosis, doctors place fingers in the three points of the patient's radial artery in the patient's radial artery, through the touch of the finger to the pulse pulsation under different pulse pressure to analyze the patient's physical condition. However, the technique of pulse diagnosis needs long-term learning and practice. In order to overcome these problems, in order to overcome these problems, in order to overcome these problems, in order to overcome these problems, more and more scholars have used the sensor technology to obtain pulse signal and use the computer technology to analyze the patient's health condition, and make the pulse diagnosis technology develop to the objective direction. The acquisition and analysis of stroke signal mainly include the acquisition of pulse signal, preprocessing, feature extraction and pattern classification. In this paper, the acquisition and analysis of pulse signal is systematically studied from these four aspects. The acquisition of pulse signal is the basis of pulse signal analysis. In recent years, different research institutions have developed various kinds of research institutions. All kinds of pulse signal acquisition system, however, the existing acquisition equipment still has some shortcomings to improve. First of all, the existing pulse signal acquisition system mostly does not have a positioning auxiliary system. The process of placing the sensor on the wrist is still largely dependent on the experience of the acquisition personnel, making the selection of the position more subjective and time-consuming. The existing system can not automatically control the sampling pressure, making the pressure setting more subjective and time-consuming. In addition, the existing equipment also has some deficiencies in the measurement of the pulse width and inch, the three parts of the scale, the ruler and the ruler. In this paper, a new pulse signal acquisition system is designed. The sampling position and the height of the probe automatically adjust to the preset pressure to reduce the subjective factors in the acquisition process and accelerate the sampling speed and improve the objectivity of the sampling. The system also has the functions of measuring pulse width and inch, Guan, and ruler three. The preprocessing of pulse signal signal is designed to remove coupling in pulse signal. Interference to improve the quality of the pulse signal and improve the accuracy of subsequent feature extraction and pattern classification. The existing preprocessing process is usually focused on removing interference from the coupled noise and baseline drift in the pulse signal, however, some interference is caused by the loss of information or the overlap of the frequency band of the pulse signal. The removal of disturbance is difficult to remove, such as saturation, and the two kinds of artifacts, such as pseudo trace, are difficult to be processed by the existing preprocessing methods. This makes the pulse data set after preprocessing to include abnormal samples, thus affecting the accuracy of feature extraction and classification. This paper extends the original preprocessing framework and adds saturation detection mode to this problem. Block and artifact detection module, and propose a difference based saturation detection algorithm and the pseudo trace detection algorithm based on the connectivity of complex network. It can effectively detect two common disturbances such as saturation and artifact. The feature extraction of pulse signal is another important lesson in pulse signal analysis. It is divided into two categories: one is a feature extraction method based on a single pulse cycle based on the analysis of the characteristics of the pulse cycle. One is a pulse feature extraction method based on the whole pulse signal based on the analysis of the whole pulse signal. The pulse signal is a quasi periodic signal, and there is a certain difference between each cycle. Different, we think this difference has a certain diagnostic significance, especially for some diseases related to rhythm. However, these two methods do not pay attention to the depiction of the difference between the pulse signal cycles. In this paper, we put forward three kinds of characteristics that are sensitive to the cycle difference from the difference between the period of the pulse signal. The difference between cycles is important for the diagnosis of disease, especially in the diagnosis of diseases that are more sensitive to periodic differences. In this paper, the characteristics obtained in this paper are better than other characteristics in the classification experiments. In the classification study, the classification performance of the pressure pulse signal and the other two more common pulse signals (the photoelectric pulse signal and the ultrasonic pulse signal) is compared, and the pulse signal fusion classification algorithm based on the combined kernel model is proposed. By analyzing the acquisition principle of the three different pulse signals, the physical meaning and the interconnected phase are analyzed. In this paper, we discuss the sensitive characteristics of different types of pulse signals. We believe that different types of pulse signals have a certain difference in the diagnosis of different diseases. If the lesion caused by a disease is related to the sensitive characteristics of a certain type of pulse signal, the pulse signal of this type of disease is in the diagnosis of this type of disease. The experimental results show that the ultrasonic pulse signal is better than the other pulse signals in the diagnosis of diabetes, and the pressure signal has achieved better diagnostic results in the diagnosis of arteriosclerosis. More effective diagnosis information is obtained step by step. Therefore, we propose a fusion classification algorithm based on combined kernel model for multiple pulse signals. In the experiment, the accuracy of diagnosis is higher than that of single pulse signal.

【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R241;TN911.6

【参考文献】