多通道混合心电信号建模及胎儿心电信号提取算法研究

发布时间：2018-03-06 18:47

本文选题：混合心电信号建模　切入点：FECG　出处：《哈尔滨工业大学》2014年硕士论文　论文类型：学位论文

【摘要】：先天性心脏缺陷形成于怀孕早期，是最常见的出生缺陷，也是致使新生儿夭折的主要原因。大部分心脏缺陷会表现在心电图中，而且心电图记录方法和常规的超声方法相比，包含了更多的生理信息。因此，定期监测胎儿心电图（FECG）及发现早期的心脏异常，可以帮助产科和儿科医生确定早期用药或在分娩过程中或出生后考虑必要的预防和诊断措施。然而胎儿心电信号十分微弱，容易受到外界环境和母体自身干扰，是目前提取较为纯净胎儿心电信号面临的主要问题。本文研究胎儿心电信号产生原理，心电噪声模型和偶极矢量模型以及多通道混合心电信号模型。重点研究上述模型的参数优化估计算法，研究模拟混合心电信号的空间滤波算法及胎儿心电信号的提取算法。主要研究内容包括以下几个方面：首先，合成心电噪声模拟信号。根据实际心电噪声模型即时变AR模型，基于卡尔曼滤波算法，并且利用真实心电噪声数据库对模型进行训练，实现模型参数优化估计，合成不同信噪比的心电噪声模拟信号，为合成多通道混合心电模拟信号以及合理优化布置母体腹部电极提供理论依据。然后，，合成多通道混合母体腹部心电模拟信号。研究基于最小二乘梯度下降算法的三维动态偶极矢量模型的参数优化，利用R波检测、心拍分片和提取特征参数实现偶极矢量模型的初始化，引入均方误差评价模型的准确度。根据多通道心电信号模型，推导出人体容积导体模型公式，利用非侵入式心电信号数据库合成多通道心电模拟信号，叠加合成的模拟噪声，输出多通道母体腹部混合心电模拟信号。接着，利用EKF和EKS算法滤除心电噪声。线性化基于极坐标形式下的五阶高斯核函数模型，建立状态方程和观测方程。利用3σ准则粗略估计两种算法的误差分布，并且引入标准信噪比，更精确的评价两种滤波算法的有效性。最后，利用盲源分离算法分离得到FECG。根据与盲源分离算法相适应的多通道心电信号（母体）模型，提取胎儿心电信号。引入修正输出信噪比对比分析Fast ICA和JADE算法的提取效果，验证两种提取算法的有效性。
[Abstract]:Congenital heart defects occur early in pregnancy, are the most common birth defects and are the leading cause of neonatal mortality. Most heart defects occur in electrocardiograms, and ECG recording methods are compared to conventional ultrasound methods. Therefore, regular monitoring of fetal electrocardiogram (FECG) and detection of early cardiac abnormalities, Can help obstetricians and paediatricians to identify early medication or to consider the necessary preventive and diagnostic measures during childbirth or after birth. However, fetal ECG signals are very weak and vulnerable to external environment and maternal interference, It is the main problem to extract pure fetal ECG signal. In this paper, the principle of fetal ECG signal generation, ECG noise model, dipole vector model and multichannel mixed ECG model are studied. The spatial filtering algorithm of analog mixed ECG signal and the extraction algorithm of fetal ECG signal are studied. Firstly, synthetic ECG noise analog signal is synthesized. According to the real ECG noise model, the AR model is changed immediately, based on Kalman filter algorithm, and the real ECG noise database is used to train the model to realize the optimal estimation of the model parameters. The synthesis of ECG noise analog signals with different signal-to-noise ratio provides a theoretical basis for the synthesis of multi-channel mixed ECG analog signals and the rational layout of the abdominal-electrode of the mother body. Then, the abdominal electrocardiogram analog signals of multi-channel hybrid matrix are synthesized. The parameter optimization of 3D dynamic dipole vector model based on least square gradient descent algorithm is studied, and R wave detection is used. The dipole vector model is initialized by taking the beat and extracting the characteristic parameters, and the accuracy of the model is evaluated by the mean square error. According to the multichannel electrocardiogram model, the formula of the human body volume conductor model is derived. Multi-channel ECG analog signals are synthesized by using non-invasive ECG database, and the synthetic analog noise is superimposed to output the mixed ECG analog signals from the abdomen of multi-channel mother. Then, EKF and EKS algorithms are used to filter ECG noise. Linearization is based on the fifth order Gao Si kernel function model in polar coordinate form, and the state equation and observation equation are established. The error distribution of the two algorithms is estimated roughly by using the 3 蟽 criterion. And the standard signal-to-noise ratio is introduced to evaluate the effectiveness of the two filtering algorithms more accurately. Finally, FECG is separated by blind source separation algorithm. According to the multi-channel ECG (matrix) model suitable for blind source separation algorithm, The effect of Fast ICA and JADE algorithm is compared with the modified output SNR. The validity of the two algorithms is verified.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.6;R714.5

【参考文献】