非侵入式胎儿心电图多域联合分离提取研究

The fetal electrocardiogram (FECG) recorded from maternal abdominal wall usually possesses an extra lower signal-to-noise ratio (SNR), which is overlapped with various kinds of interferences and noises in time and frequency domain to a great extent. As a result, it remains an open problem in biomedical signal processing field to completely extract the fetal ECG morphology from the abdominal ECG signal. In this project, the priori knowledge of signal and noise such as their rules of movement, change and development is deeply considered to look for their separable or partially separable feature space. By combining multiple separable feature spaces, a multi-domain jointed layering processing method is proposed to step by step separate the interferences and noises mixed with FECG. First, based on the modeling and estimating theory of stochastic system, a multi-scale sequential dynamic estimation method is proposed to separate MECG in the time-scale domain using its dynamic evolution law and multi-scale information comprehensively. Second, treating the fetal heart as a cyclical beating mechanical device, a novel cyclo-stationary semi-blind separation algorithm is presented to separate the random impulsive noise in the space-statistical domain on the basis of cyclo-stationary feature extraction theory of the rotating machinery vibration signal. Finally, the FECG can be completely extracted by suppressing the broad-band low-amplitude noise.According to the above, there is no doubt that this study will lay a solid theoretical and technical foundation for promoting the fetal heart rate monitoring to fetal electrocardiogram monitoring.

非侵入式探测的胎儿心电图(FECG)，信噪比极低，与多种干扰噪声在时频域严重重叠，目前尚无完整分离提取FECG的有效方法，严重制约了胎儿监护水平。本课题充分利用信号与干扰噪声的先验知识，分析研究其运动、变化、发展规律，寻找它们可分的特征空间，从多个可分特征空间联合处理入手，提出一种多域联合分层处理方法，逐层分离掩盖在FECG上的各类干扰噪声，完整保留FECG：首先从随机系统建模与估计理论出发，综合利用母体心电(MECG)的动态演化规律及其多尺度信息，提出一种多尺度贯序式动态估计方法，从时间-尺度域分离MECG；其次，把胎儿心脏看作一种周期性搏动的机械装置，从旋转机械振动信号循环平稳特征提取理论出发，探索以循环平稳特性为先验信息的FECG半盲分离方法，从空间-统计域分离随机脉冲噪声；最后分离宽带低幅噪声，实现FECG完整提取，为推进围产期胎儿心率监护到胎儿心电监护的跨越奠定理论与技术基础。

胎儿心电图是认知和了解胎儿心脏活动的常规手段之一，对胎儿疾病的早期发现意义重大。然而目前母体腹壁胎儿心电图提取与辨识存在相当的难度，且胎儿心电波形结构缺乏标准描述。为此，本课题围绕母体腹壁胎儿心电数据采集、信噪分离、胎心率检测等展开深入研究：①针对腹壁混合信号中，母体心电干扰强度最大、时域波形结构具有准周期的特点，研究构造关于母体心拍的周期矩阵，进而提出一种基于单通道奇异值分解（SVD）的母体心电干扰消除算法；②针对剩余混合信号中，胎儿心电信号特有的循环平稳特性，提出了一种基于循环频率的多通道胎儿心电半盲提取方法，有效分离残留母体心电与肌电等脉冲型干扰噪声；③针对上述多导联腹壁心电采集方法难以应用于家庭在线胎儿监护问题，提出一种新的低复杂度单通道胎心率鲁棒估计算法，算法对电极配置要求的简便性以及算法本身的递归性，使得在客户端本地实现胎儿心率实时估计成为可能，适用于胎儿心率的家庭动态监护环境；④在此基础上，进一步研制开发了基于胸带式心电采集器的可穿戴母体腹壁心电采集系统，并展开初步的数据测试试验。