HRV分析在心衰诊断和新生儿疼痛检测中的应用研究

发布时间：2018-04-21 21:37

本文选题：心率变异性 + R波检测　；参考：《中南大学》2014年博士论文

【摘要】：摘要：连续窦性心拍之间的时间间隔存在微小涨落,这种现象称为心率变异性(Heart Rate Variability, HRV)。 HRV蕴藏了丰富的生理和病理信息,是评估自主神经系统功能的一个重要标志。HRV分析在疾病诊断、情绪识别和脑力负荷评估等诸多领域有着广泛的应用。本文对HRV信号的获取、分析和应用中的一些相关问题进行了研究。论文的主要研究内容如下： 1)提出了基于小波系数模极大值序列跃变点的R波检测策略,实现了连续小波变换对心电信号R波的检测。利用复Morlet小波与Mexican-hat小波对心电信号进行连续小波变换后,小波系数模极大值对应R波峰值的特点,通过基于小波系数模极大值序列跃变点的R波检测策略,在上述两种小波系数的线性组合中检测R波,平均灵敏度为99.37%,平均阳性预测率为99.35%。 2)提出了基于CEEMD分解的RR间期序列去趋势方法。从心电信号中提取的RR问期序列是HRV分析的信息来源,并且是非均匀采样的。为了得到准确的HRV分析结果,需要在预处理阶段将RR间期序列中缓慢的趋势予以去除。平滑先验方法(Smoothness Prior Approach, SPA)目前使用最为广泛,但这一方法需要将非均匀采样的RR间期序列通过重采样转换为均匀采样序列。这一过程将产生噪声,并使信号的质量受到损害。为了解决这一问题,引入了经验模态分解(Empirical Mode Decomposition, EMD)。将分解后的信号通过部分重构,去除其趋势成分。这一方法可直接用于非均匀采样信号的处理。此外,为了能够采用标准指标评价去趋势方法的性能,提出了一个RR间期序列模型。采用以分贝计的信噪比(ISNR)、均方误差(EMS)和百分比均方根误差(DPRS)评价RR间期序列的去趋势性能。结果表明,与SPA方法相比,基于互补整体EMD(Complementary Ensemble EMD, CEEMD)的去趋势方法能得到更高的ISNR,更低的EMS和DPRS,说明其具有更好的性能,并能由此得到更准确的HRV分析结果。 3)比较了心衰病人和健康人的HRV指标,并建立了基于相关指标的心衰诊断模型。采用时域、频域和非线性方法对40名健康人和40名心衰病人的心电数据进行了短时HRV分析,从而建立了基于不同指标组合和线性判别分析(Linear Discriminant Analysis, LDA),及支持向量机(Support Vector Machine, SVM)的心衰诊断模型。结果表明,基于RR间期均值RR、RR间期标准差SDNN、去趋势波动分析(Detrended Fluctuation Analysis, DFA)短期波动斜率α1、DFA长期波动斜率α2、近似熵ApEn等5个指标和LDA的诊断模型诊断正确率可达到92.5%；基于RR、SDNN、RR间期差值的均方根RMSSD、Poincare分析短轴参数SD1、ApEn等5个指标和SVM的模型诊断正确率可达到95%。HRV的相关指标可揭示心脏的动力学特征,并可用于心衰的诊断。 4)研究了足跟取血造成的疼痛暴露对新生儿自主神经系统的影响,并建立了基于HRV指标组合的新生儿疼痛检测模型。采用时域、频域和非线性方法对40名新生儿疼痛暴露前后心电数据进行了短时HRV分析,并建立了基于不同指标组合和LDA,及SVM的疼痛检测模型。结果表明,基于ApEn、递归图分析最大对角线长度Lmax、确定性DET等3个指标和LDA的新生儿疼痛检测模型检测正确率达到78.75%,基于RR、相邻两个RR间期对差值大于50ms的百分比pNN50、ApEn、关联维D2、递归率REC等5个指标和SVM的模型检测正确率达到83.75%。HRV的相关指标可反映新生儿自主神经系统对疼痛暴露的应答,相关指标的组合可用于新生儿疼痛检测。
[Abstract]:Abstract: there are small fluctuations in the time interval between continuous sinus racket. This phenomenon is called Heart Rate Variability (HRV). HRV contains abundant physiological and pathological information. It is an important marker for evaluating the function of autonomic nervous system, which is an important marker of.HRV analysis in the diagnosis of disease, emotion recognition and brain load assessment. The domain is widely applied. This paper studies the acquisition, analysis and application of HRV signals. The main contents of the paper are as follows:
1) a R wave detection strategy based on the jump point of the wavelet coefficient modulus maximum sequence is proposed to detect the R wave of the ECG signal by continuous wavelet transform. After the continuous wavelet transform between the complex Morlet and Mexican-hat wavelets, the maximum value of the wavelet coefficient modulus corresponds to the peak value of the R wave, and is based on the modulus maximum of the wavelet coefficients. The R wave detection strategy of the value sequence jump point detects the R wave in the linear combination of the above two wavelet coefficients, the average sensitivity is 99.37%, and the average positive predictive rate is 99.35%.
2) the RR interval sequence detrending method based on CEEMD decomposition is proposed. The RR query sequence extracted from the ECG signal is the source of the HRV analysis information and is nonuniform sampling. In order to obtain accurate HRV analysis results, the slow trend in the RR interval need to be removed at the preprocessing stage. The smooth prior method (Smoothness Pr) is needed. IOR Approach, SPA) is currently the most widely used, but this method needs to convert the non uniform sampled RR interval sequence into the uniform sampling sequence through resampling. This process will produce noise and damage the quality of the signal. In order to solve this problem, the empirical mode decomposition (Empirical Mode Decomposition, EMD) will be introduced. The signal is partially reconstructed to remove its trend component. This method can be used directly for the processing of nonuniform sampling signals. In addition, a RR interval sequence model is proposed in order to evaluate the performance of the detrend method with standard index. The mean square error (EMS) and the root mean square error of the signal to noise ratio (ISNR), the mean square error (EMS) and the percentage mean square error are adopted. (DPRS) the detrending performance of the RR interval is evaluated. The results show that the detrending method based on the complementary integral EMD (Complementary Ensemble EMD, CEEMD) can get higher ISNR, lower EMS and DPRS, indicating that it has better performance and can get more accurate HRV analysis results from this method compared with the SPA method.
3) the HRV index of patients with heart failure and healthy people was compared, and a diagnostic model of heart failure based on related indexes was established. The time domain, frequency domain and nonlinear methods were used to analyze the ECG data of 40 healthy people and 40 heart failure patients by short time HRV analysis, which was based on the combination of different indexes and linear discriminant analysis (Linear Discriminant Anal). Ysis, LDA) and the diagnosis model of heart failure of Support Vector Machine (SVM). The results show that, based on the RR interval mean RR, RR interval standard deviation SDNN, detrending fluctuation analysis (Detrended Fluctuation), short-term wave slope alpha 1, long wave slope alpha 2, approximate entropy, and other 5 indexes and diagnostic model diagnosis The accuracy can reach 92.5%; the root mean square (RMS) RMSSD based on RR, SDNN and RR interval values, the Poincare analysis of the short axis parameter SD1, ApEn and so on, and the correct rate of the model diagnosis of SVM can reach 95%.HRV, which can reveal the dynamic characteristics of the heart, and can be used for the diagnosis of heart failure.
4) the effects of pain exposure on the heel extraction on the autonomic nervous system of the newborn were studied, and a neonatal pain detection model based on HRV index combination was established. The time domain, frequency domain and nonlinear methods were used to analyze the ECG data of 40 neonates before and after pain exposure, and a combination of different indexes and LDA was established. And SVM's pain detection model. The results show that, based on ApEn, recursive graph analysis of the maximum diagonal length Lmax, deterministic DET and other 3 indicators, and LDA for neonatal pain detection model detection accuracy reached 78.75%, based on RR, two adjacent RR intervals are more than 50ms in the percentage pNN50, ApEn, associated dimension D2, recursion REC, and other 5 indicators. The correlation index of the correct rate of 83.75%.HRV can reflect the response of the neonatal autonomic nervous system to pain exposure, and the combination of the related indicators can be used for neonatal pain detection.

【学位授予单位】：中南大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R722.1;R541.6

【参考文献】