A Non-Invasive Fetal Electrocardiogram Extraction Algorithm Based on ICA Neural Network

Yalan Ye, Quanyi Mo, Zhiling Zhang

Abstract: The extraction of fetal electrocardiogram (ECG) from maternal skin electrode measurements is an open problem in recent decades. Many researchers proposed blind source separation (BSS) or Independent Component Analysis (ICA) based neural network methods to address this problem. However, by these methods all of the source signals are simultaneously separated, but in fact only one source signal is the desired FECG and others are unwanted ones. In contrast, blind source extraction (BSE) only outputs a single source and is closely related to BSS, which is obviously a better choice. In this paper, we propose a non-invasive extraction algorithm based on ICA neural network that can extract the desired FECG with little noise as the first extracted signal. The algorithm is very robust to outliers. The real-data world has shown that the algorithm can achieve satisfying results.