基于半盲反卷积源分离的胎儿心电信号提取研究

发布时间：2019-03-21 13:22

【摘要】：胎儿心电图是目前最为有效的围产期胎儿监护手段之一。医生通过分析胎儿心电波形变化及早发现胎儿在宫内发育时产生的病变情况,从而降低初生婴儿损伤率和死亡率。但非侵入式法采集的孕妇腹壁混合心电信号中,胎儿心电信号信噪比低且常受母体心电和其他强噪声干扰,因此如何提取清晰的胎儿心电信号一直是胎儿监护的重要研究课题。盲源分离(BSS)是目前最具研究前景的胎儿心电信号提取算法,然而现有基于BSS的胎心电提取算法通常采用线性瞬时混合模型且未充分利用心电信号特征信息,导致提取结果依然存在准确率低且生理学意义模糊等问题。为此,本论文结合信号特性和混合模型提出了一种基于卷积模型的半盲胎儿心电信号提取算法。具体内容如下:(1)分析胎儿心电信号特性并构建线性卷积混合模型。由电生理特性入手进行胎儿心电信号特性分析,利用非最小相位特性验证了采用盲反卷积法提取胎儿心电信号的合理性;结合实际心电信号特性和盲反卷积源分离混合模型,对腹壁心电信号构建线性卷积混合模型并给出基于心电特性的胎儿心电信号提取方法。(2)基于循环平稳特性和信号卷积混合模型,提出了半盲反卷积源分离算法。半盲反卷积单源分离算法将时延和循环频率参数引入到基于非高斯性度量的目标函数构造中,并通过梯度法最大化目标函数实现了母胎心电单源信号的有效分离。为提高胎儿心电信号估计准确度,利用最小二乘逆滤波母体心电消除算法改进传统BSS中的时域相减法。实验仿真表明,在母胎时域波形重叠情况下,最小二乘逆滤波母体心电消除算法较之传统时域相减法在胎儿心电波形保留完整性方面效果更好。(3)基于半盲反卷积源分离算法原理,提出了完整的胎儿心电信号提取方法。结合半盲反卷积单源分离和母体心电消除算法,提出了包括混合心电信号去噪预处理、特征参数估计和胎心电信号提取增强三部分的胎儿心电信号提取方法。分析实验提取结果并将半盲反卷积源分离法与传统BSS进行对比,实验结果显示,利用半盲反卷积源分离法提取的胎儿心电信号与实验数据库胎儿头皮电极采集的胎儿心电参考信号在胎心率和波形均拍方面极为接近,并且在可视化波形和信噪比等性能评价指标方面优于传统BSS。
[Abstract]:Fetal electrocardiogram is one of the most effective perinatal fetal monitoring methods. By analyzing the changes of fetal ECG waveform, we can find out the pathological changes of fetus during intrauterine development as early as possible, so as to reduce the rate of injury and mortality of newborn infants. However, the signal-to-noise ratio of fetal ECG signal is low and often interfered by maternal ECG and other strong noises in the mixed ECG signals collected by non-invasive method in pregnant women's abdominal wall. Therefore, how to extract clear fetal ECG signals has always been an important research topic in fetal monitoring. Blind source separation (BSS) is the most promising fetal ECG signal extraction algorithm at present. However, the existing BSS-based FECG extraction algorithms usually use the linear instantaneous mixed model and underutilize the ECG signal feature information. As a result, there are still some problems, such as low accuracy and vague physiological meaning. In this paper, a semi-blind fetal ECG signal extraction algorithm based on convolution model is proposed based on signal characteristics and mixed model. The main contents are as follows: (1) analyze the characteristics of fetal ECG signal and construct the linear convolution mixed model. By analyzing the electrophysiological characteristics of fetal ECG signals, the rationality of using blind deconvolution method to extract fetal ECG signals was verified by using non-minimum phase characteristics. Combining the characteristics of actual ECG signals and the mixed model of blind deconvolution source separation, The linear convolution hybrid model of abdominal ECG signal is constructed and the method of fetal ECG signal extraction based on ECG characteristics is presented. (2) based on the mixed model of cyclostationary and signal convolution, a semi-blind deconvolution source separation algorithm is proposed. Semi-blind deconvolution single-source separation algorithm introduces delay and cyclic frequency parameters into the construction of objective function based on non-Gaussian metric, and achieves the effective separation of single-source signal of maternal fetal ECG by maximizing the objective function of gradient method. In order to improve the accuracy of fetal ECG signal estimation, the least square inverse filtering algorithm is used to improve the time domain subtraction method in traditional BSS. The simulation results show that when the waveform of the mother-fetus overlaps in time domain, The least square inverse filtering algorithm is better than the traditional time domain subtraction method in preserving the integrity of fetal ECG waveform. (3) based on the principle of semi-blind deconvolution source separation algorithm, A complete method of fetal ECG signal extraction was proposed. Combined with semi-blind deconvolution single source separation and maternal ECG cancellation algorithm, a new method of fetal ECG signal extraction is proposed, which includes mixed ECG signal de-noising pre-processing, feature parameter estimation and fetal ECG signal extraction and enhancement. The experimental results were analyzed and the semi-blind deconvolution source separation method was compared with the traditional BSS. The experimental results showed that The fetal ECG signal extracted by the semi-blind deconvolution source separation method is very close to the fetal ECG reference signal collected by the fetal scalp electrode in both fetal heart rate and waveform. And it is better than traditional BSS. in visual waveform and signal to noise ratio.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：R714.5;TN911.7

【相似文献】