An improved parallel sub-filter adaptive noise canceler for the extraction of fetal ECG
An improved parallel sub-filter adaptive noise canceler for the extraction of fetal ECG
复制标题
DOI：
10.1515/bmt-2020-0313
复制
发表时间：
2021-10-01
影响因子：
1.7
通讯作者：
Kumar, R.
中科院分区：
文献类型：
--
作者：
Krupa, Abel Jaba Deva;Dhanalakshmi, Samiappan;Kumar, R.
无创提取胎儿心电图（FECG）通过处理腹部信号是新兴的一种有前途的方法，在妇产科领域。本文提出了一种两级改进的非线性自适应滤波方法。首先使用自适应神经模糊推理系统（ANFIS）对自适应噪声消除器（ANC）的参考输入进行处理，以估计腹部信号中的非线性母成分。提出了一种平行子滤波器（PSF） ANC来评估胎儿腹部信号。PSF-ANC将单个自适应滤波器分解成多个子滤波器以提高收敛性能。采用归一化最小均方算法，通过使均方误差最小化，自适应地获得了PSF-ANC的滤波系数。基于误差信号的计算，提出了不同的误差算法和常见的误差算法。利用FECG综合数据库中的综合数据对收敛性能进行了评价。利用来自Daisy数据库和Physionet非侵入性FECG数据库的两个实时数据，对所提出的anfiss - psf的性能进行定性和定量评估。结果证明，与现有技术相比，所提出的anfi - psf ANC的性能有所提高。该方案的灵敏度为97.92%，准确率为94.52%，阳性预测值为94.66%，F1评分为96.12%。
Non-invasive extraction of fetal electrocardiogram (FECG) by processing the abdominal signals is emerging as a promising approach in the areas of obstetrics and gynecology. This paper presents a two-stage improved non-linear adaptive filter for FECG extraction. The reference input to the adaptive noise canceler (ANC) is first processed using an adaptive neuro-fuzzy inference system (ANFIS) to estimate the non-linear maternal component in abdominal signals. A parallel sub-filter (PSF) ANC is proposed to assess the fetal ECG from the abdominal signal. The PSF-ANC decomposes a single adaptive filter into multiple sub-filters to improve the convergence performance. The filter coefficients of PSF-ANC adaptively obtained using normalised least mean square algorithm by minimizing the mean square error. Different error and common error algorithms are proposed based on the computation of the error signal. A synthetic data from the FECG synthetic database is used to evaluate the convergence performance. Two real-time data from the Daisy database and the Non-invasive FECG database from Physionet are used to evaluate the proposed ANFIS-PSF's performance qualitative and quantitatively. The results justify the performance improvement of proposed ANFIS-PSF ANC compared to the state of art techniques. The proposed scheme achieves a sensitivity of 97.92%, 94.52% accuracy, a positive predictive value of 94.66%, and an F1 score of 96.12%.