Foetal ECG extraction from maternal body surface measurement using independent component analysis

Abstract

A method applying independent component analysis (ICA) to detect the electrocardiogram of a prenatal cattle foetus is described. Three channels of signal, one is from chest lead and two are from abdominal leads, are picked up noninvasively by attaching disposable cutaneous electrodes on the body surface of a maternal cow. Measured signals are the mixtures of components including maternal ECG, foetal ECG and random noise. Such mixture procedure is expressed by a time-invariant linear model. To separate foetal ECG from maternal ECG and random noise, the ICA method is applied to rind an optimal separating matrix only using measured signals. The optimization uses hyperbolic tangent as a contrast function and minimizes it based upon the principle of mutual information. The method was examined by simulation signals generated by random mixture of two-signal and one-noise sources. Real signals, measured from a pregnant cow having 177-day gestation, are used to verify the separation. The results show that in simulation, both signal and one noise are clearly separated. Real measurement is successfully distinguished into three signal sources, maternal ECG, foetal ECG and random noise. The results suggested the effectiveness of ICA approach in detecting foetal ECG from maternal body surface measurement.