Ahmadieh Hajar, Asl Babak Mohammadzadeh
Department of Electrical and Computer Engineering, Tarbiat Modares University, Tehran, Iran.
Department of Electrical and Computer Engineering, Tarbiat Modares University, Tehran, Iran.
Comput Methods Programs Biomed. 2017 Apr;142:101-108. doi: 10.1016/j.cmpb.2017.02.009. Epub 2017 Feb 22.
We proposed a noninvasive method for separating the fetal ECG (FECG) from maternal ECG (MECG) by using Type-2 adaptive neuro-fuzzy inference systems.
The method can extract FECG components from abdominal signal by using one abdominal channel, including maternal and fetal cardiac signals and other environmental noise signals, and one chest channel. The proposed algorithm detects the nonlinear dynamics of the mother's body. So, the components of the MECG are estimated from the abdominal signal. By subtracting estimated mother cardiac signal from abdominal signal, fetal cardiac signal can be extracted. This algorithm was applied on synthetic ECG signals generated based on the models developed by McSharry et al. and Behar et al. and also on DaISy real database.
In environments with high uncertainty, our method performs better than the Type-1 fuzzy method. Specifically, in evaluation of the algorithm with the synthetic data based on McSharry model, for input signals with SNR of -5dB, the SNR of the extracted FECG was improved by 38.38% in comparison with the Type-1 fuzzy method. Also, the results show that increasing the uncertainty or decreasing the input SNR leads to increasing the percentage of the improvement in SNR of the extracted FECG. For instance, when the SNR of the input signal decreases to -30dB, our proposed algorithm improves the SNR of the extracted FECG by 71.06% with respect to the Type-1 fuzzy method. The same results were obtained on synthetic data based on Behar model. Our results on real database reflect the success of the proposed method to separate the maternal and fetal heart signals even if their waves overlap in time. Moreover, the proposed algorithm was applied to the simulated fetal ECG with ectopic beats and achieved good results in separating FECG from MECG.
The results show the superiority of the proposed Type-2 neuro-fuzzy inference method over the Type-1 neuro-fuzzy inference and the polynomial networks methods, which is due to its capability to capture the nonlinearities of the model better.
我们提出了一种利用二类自适应神经模糊推理系统从母体心电图（MECG）中分离出胎儿心电图（FECG）的非侵入性方法。
该方法可通过使用一个腹部通道（包括母体和胎儿心脏信号以及其他环境噪声信号）和一个胸部通道，从腹部信号中提取FECG成分。所提出的算法检测母体身体的非线性动力学。因此，从腹部信号中估计出MECG的成分。通过从腹部信号中减去估计出的母体心脏信号，可提取出胎儿心脏信号。该算法应用于基于McSharry等人和Behar等人开发的模型生成的合成心电图信号，以及DaISy真实数据库。
在高不确定性环境中，我们的方法比一类模糊方法表现更好。具体而言，在基于McSharry模型的合成数据评估算法时，对于信噪比为-5dB的输入信号，与一类模糊方法相比，提取的FECG的信噪比提高了38.38%。此外，结果表明，增加不确定性或降低输入信噪比会导致提取的FECG的信噪比提高百分比增加。例如，当输入信号的信噪比降至-30dB时，我们提出的算法相对于一类模糊方法将提取的FECG的信噪比提高了71.06%。基于Behar模型的合成数据也得到了相同的结果。我们在真实数据库上的结果反映了所提出的方法成功分离母体和胎儿心脏信号，即使它们的波形在时间上重叠。此外，所提出的算法应用于具有异位搏动的模拟胎儿心电图，在从MECG中分离FECG方面取得了良好的结果。
结果表明，所提出的二类神经模糊推理方法优于一类神经模糊推理和多项式网络方法，这是由于其能够更好地捕捉模型的非线性特性。
