自适应滤波算法在心电监护中的应用研究/

2019-04-28 08:41:33

The noise 噪声 ECG 心电监护



本课题的研究提出了一种能够有效提取淹没在强噪声中的心电信号的算法。经实验证明该算法运算快速、高效,在心电监护系统中具有重要的研究意义。
心电信号(ECG singal)是人体最重要的电生理参数之一,心电信号随着时间及检测状态的变化,通常具有非平稳的特点。心电信号的检测和处理在生物医学信号的研究中具有十分重要的意义。通过安放在身体表面的胸电极或四肢的电极,将心脏产生的电位变化以时间为函数记录下来的曲线就称为心电图(electro- cardiogram, ECG)。将心电信号从人体记录到心电图的通道成为导联。心电信号是一个向量,所以体表测得的心电图会由于位置的不同而具有不同的特征,但基本上都包括一个P波、一个QRS波群和一个T波,有时在T波之后还出现一个小的U波。心电图作为心脏电活动的图形记录,是心脏疾病诊断中起决定性作用的诊断手段。
心电监护就是实时的监测病人的心电波形或通过心电图回放,对心电波形进行分析、给出诊断结果,判断病人的具体情况。心电监护技术对于心脏病人的早期发现和治疗、病例跟踪反馈检查和医学研究都有举足轻重的作用。心电监护的首要任务就是心电图波形的检测。在心电监护过程中,由于环境及患者自身状况的变化,难免会引入噪声,噪声将会严重影响到心电监护的自动分析功能。可能引入的噪声主要分为以下四类:工频(50Hz/60Hz)干扰;心电以外人体电现象所引起的噪声;基线漂移;强噪声。工频干扰主要由室内的照明及动力设备影响到人体的分布电容所引起;心电以外的电现象主要指肌电以及胎儿心电对母亲心电的影响;基线漂移是由于人体的移动、呼吸或是电极接触不良等情况都会造成ECG 信号偏离原来的基线水平而发生漂移的现象;产生强噪声的原因有很多,例如临床手术时高频电刀的使用、电极的松动、Holter监护病人周围大型电器的开关等等。四类噪声中,强噪声的表现形式最多,对ECG的影响也最大。目前国内外针对前三种噪声的处理已经形成了较成熟的理论和实际应用,由于强噪声的频带非常宽,与心电信号的频带重叠,而且幅度很高,常将心电信号完全淹没。尽管前人已经采用了一些软、硬件方法对抗强噪声干扰进行了一系列的研究,但效果并不理想。因此研究一种能够有效抑制心电监护中强噪声的方法对于心电监护具有十分重要的意义。
本课题所作的工作就是针对心电监护中遇到的强噪声干扰,通过多导联综合计算,根据Widrow提出的自适应噪声消除器来消除强噪声在心电图中的影响。经典的信号处理方法是基于对干扰或噪声的概率分布函数作理想化假定,在假定的理想噪声模型下设计对某些性能准则为最优的信号处理器(如维纳滤波器、匹配滤波器等)来处理这类信号。然而,在实际应用中遇到的随机信号的统计特征通常是未知的,处理这种情况的一种解决方法就是使用自适应噪声消除。自适应噪声消除器有基本输入和参考输入两个输入端。基本输入由携带信息的有用信号和与其不相关的干扰组成,而参考输入为与输入端干扰的相关形式。参考输入将通过基于最小均方误差(LMS)算法进行抽头权值自适应的横向滤波器,该滤波器使用参考输入对包含在基本输入端的干扰信号进行估计,从基本输入中减去自适应滤波器的输入,即可消除噪声的影响。实际研究研究中,我们将含有强噪声的心电数据作为基本输入。根据导联网络理论,通过多导联的综合运算提取出的噪声的相关形式作为参考输入。
本文的技术关键有三处。(1)参考导联的选择。有自适应噪声消除器的原理知道,参考输入必须是基本输入中噪声的相关形式,因此在多导联综合计算时,应当选取适当的导联参与运算,以保证计算的最终结果是噪声的相关形式,同时与心电信号不相关或弱相关。(2)滤波器抽头向量初始权值的设置。自适应的过程就是滤波器的权值不断改变的过程。由于心电信号自身的周期性,随着自适应过程的进行,滤波器权值逐渐趋于稳定,收敛于某一固定的值或在其周围一个很小的范围内波动。通常在没有先验知识的情况下,初值都为零。本文通过大量实验对滤波器权值进行统计,在滤波开始时对滤波器权值进行合理的初始化设置,使滤波过程具有更快的收敛速度。(3)步长因子的调节。由LMS算法理论可知,步长因子µ应满足: ,其中Smax是抽头输入输入的功率谱密度最大值,M为滤波器长度。这是自适应过程收敛的充分必要条件。由于强噪声的随机性很大,我们设置的步长因子初始值并不一定能够保证自适应滤波过程始终收敛。因此,我们在分块处理过程中,同时用直接法预测下一块的功率谱,根据预测的功率谱来调整步长因子µ,以保证滤波过程中处处收敛。
本文的创新之处在于,在现有的心电监护条件下,不增加参考电极等辅助设备,从算法上寻求解决心电监护受到强噪声干扰问题的方法,并取得了不错的效果。
实验结果表明,通过多导联的综合运算,采用自适应滤波系统,有效的滤出了心电信号中的强噪声,有效的改善了ECG 波形的显示,有利于ECG 的精确分析和诊断。同时,本文提出的算法运算速度快、准确率较高,具有一定深入研究的意义和推广价值。




An efficient algorithm for abstract ECG signals from strong-noise is presented in this research. The algorithm was proved to be fast and high efficient. It has important meaning in ECG monitoring.
ECG signal is one of the most important electrophysiological parameters. ECG signal is non-stationary as the environment and detecting condition changed. The detecting and processing of ECG signal has very important meaning in research on bio-medical engineering signal-processing. The curve records ECG signal intitules ECG. The channels transfer ECG signal intitules leads. ECG signal is a vector, so ECG gotten from different positions has different characters. Mostly ECG has a P-wave ,a QRS-complex and a T-wave, sometimes a little U-wave appear after T-wave. ECG as the record of heart activity plays a decisive role in diagnoses of heart disease.
ECG monitoring is to analyse ECG and diagnose the patient through real-time inspecting or review the ECG. ECG means very import to heart disease patient and medical research. The first task of ECG monitoring is to detect the figure of ECG.
Noises enter ECG as the environment and the patient changed. Noises will infect the autoanalysis function of the ECG monitor. Noises in ECG can mainly partition into four kinds: power frequency interference (50/60Hz), electorphysiological signals except ECG signal, baseline drift and strong noise. Power frequency interference is produced by lighting and power impetus. Electrophysiological except mainly means electromyography(EMG) and ECG of fetus. Baseline drift is produced by body move, breath and bad electrode contract. Electrosurgical unit (ESU) and large electrical equipment etc can generate strong noise. Strong noise has wide frequency bands and high amplitude and it often submerge ECG signal. There is no good method to remove strong noises presently neither software nor hardware, so research an effective method to restrain strong noise is important and necessary for ECG monitoring.
The aim of the research is to find a method to remove strong noise from ECG based on multiple leads calculator and adaptive noise eliminator. In actual application, we don’t know the statistical characteristics of mostly random signals. Adaptive noise eliminator is used to resolve the problem. Adaptive noise eliminator has two inputs. The base input comprises useful signal and noise. The referenced input was noise correlated with the noise in base input. In actual research , ECG including strong noise is the base input and the noise calculated from multiple leads is the reference input.
The research has three key techniques. (1) Choice of leads. The leads we chosen should ensure the referenced input to be correlated with the noise in basic input. (2) Setting of weighted vector initializing. The weighted vector will gradually converge at aptotic value. We initialize some special value on weighted vector based on statistical data. It will accelerate the processing of convergence. (3) Adjusting of step size. Step is restricted by formula: . In filtering we forecast the power spectral of next segment to adjust step size to meet the adaptive process convergence.
The innovation of research is that we filter the strong noise from ECG without any additional equipment.
Experiment results proved that we abstract ECG signal from strong noise efficiently. The figure of ECG improved through multiple leads calculation and using adaptive filter. The algorithm calculates fast and efficiently. It has value to significance to research deeply and generalize widely.