一种基于单通道腹部信号的胎儿心电提取算法

设计一种基于单通道孕腹部信号的胎儿心电提取算法,分别提取出母亲心电和胎儿心电,并计算出母亲心率和胎儿心率。首先对单通道孕腹部信号进行k-TEO(k=19)变换,突出母亲心电的QRS波,从而通过简单的阈值法确定母亲心电的R波位置,接着通过在相邻R波间重采样以获得相同的R-R间期T,这样经过一个间隔为T的梳状滤波器就可以分离出相同R-R间期的母亲心电,然后再一次在相邻R波间进行重采样恢复原来的R-R间期就可以获得实际的母亲心电了。原始腹部信号减去上面提取的母亲心电后,胎儿心电QRS波的信噪比大大提高,通过再次应用提取母亲心电的算法即可得到“干净”的胎儿心电波形。选取Physionet数据库中的8 组(26 通道)孕腹部信号数据进行分析,计算每个通道数据的胎儿心电QRS波位置识别灵敏度、阳性检测率和准确性。结果表明,胎儿心电QRS波的识别准确率达到87.1%,其中有6 个通道达到100%。另外计算每个通道的母亲心率和胎儿心率并做统计分析,发现每一组中各个通道的母亲平均心率和胎儿平均心率都非常接近,同一组中各通道间母亲平均心率最大误差为0.1次/min, 而胎儿平均心率最大误差也只有0.9次/min,进一步证明算法的可靠性。

A Synthetic Algorithm of FECG Extraction from Single-Lead Abdominal Signals

In this paper, a synthetic algorithm was proposed to extract the fetal electrocardiogram (FECG) from single-lead abdominal signals. We extracted maternal electrocardiogram (MECG) and FECG successively and then calculated maternal and fetal heart rate. In the algorithm, we first applied Teager energy operator with parameter k (k=19) to protrude QRS waves of MECG, so that the location of maternal R peaks was detected correctly by simple threshold. Then we resampled between every adjacent R peaks to get same R-R interval T. After applying a comb filter whose teeth coincide with the maternal R-R interval T, we obtained the MECG with same R-R interval. And the real MECG was obtained by resampling between every adjacent R peaks in the filtered signals again to recover the original R-R intervals. The extracted MECG was subtracted from the abdominal signals. In the residual signal the QRS waves of FECG had better signal-noise-ratio. Next, the same procedure was applied on the residual signal to extract the FECG signal. We analyzed 8 sets of abdominal signals (total 26 channels) that were selected from the Physionet non-invasive FECG database. The sensitivity (Se), positive predictive value (PPV) and accuracy (F₁) of fetal QRS waves detection were calculated. Results showed that the detection accuracy of fetal QRS waves reached 87.1%, with six channel even reached 100%. The maternal heart rate (MHR) and fetal heart rate (FHR) of each channel were also calculated. We found out that the MHRs, as well as the FHRs, showed good consistence with channels of the same set. In the same set, the maximum error of the mean MHR was 0.1 times per minute and the maximum error of mean FHR was 0.9 times per minute. Such a phenomenon also proved the reliability of the proposed algorithm.