Identification of FECG from AECG Recordings using ICA over EMD

Extraction of fetal ECG (fECG) signal is essential for monitoring the health of fetus during pregnancy and helps in early diagnosis of heart abnormalities, which leads to increased infant mortality rate and post-natal complications. In real scenarios, extraction of clear fECG is challenging due to maternal ECG (mECG) and other contaminated noise (such as: baseline wander and high frequency noise). This paper is focused on design, implementation, and verification of a robust approach for fECG extraction, recorded by non-invasive procedure from the pregnant women, using empirical mode decomposition (EMD), independent component analysis (ICA), and FIR filtering. The combined EMD and ICA approach are found suitable for effective extraction in real and synthetic data. EMD separates the non-stationary and non-linear time varying signals like ECG into various modes, having high to low frequencies using signal itself as a basis. The coefficients obtained during this decomposition are called intrinsic mode functions (IMFs) representing various frequency components. Different number of IMFs are combined with the residuals to create the data matrix (or mixed signals), which are fed to the ICA (extended efficient Fast-ICA and multi-combi ICA) for separating the independent components (ICs) due to their strength in separating the combination of various distribution signals. These extracted ICs (such as: thorax ECG, fECG, and noises etc.,) are subjected to FIR filtering to obtain the fECG and its corresponding heart rate (HR). This technique is validated on simulated signals for separation, prior to applying on fECG synthetic-data and aECG-data collected from PhysioBank ATM. The performance of ICA algorithm is evaluated by API.