Cyclostationarity-Based Estimation of the Foetus Subspace Dimension from ECG Recordings

In this work, a novel method based on the cyclostationary properties of electrocardiogram (ECG) signals is introduced in order to classify independent subspaces into components reflecting the electrical activity of the foetal heart and those corresponding to mother’s heartbeats, while the remaining ones are mainly due to noise. This research is inspired from multidimensional independent component analysis (MICA), a method that aims at grouping together into independent multidimensional components blind source separated signals from a set of observations. Given an input set of observations, independent component analysis (ICA) algorithms estimate the latent source signals which are mixed together. In the case of ECG recordings from the maternal thoracic and abdominal areas, the foetal ECGs (FECGs) are contaminated with maternal ECGs (MECG), electronic noise, and various artifacts (respiration, for example). When ICA-based methods are applied to these measurements, many of the output estimated sources have the same physiological origin: the mother’s or the foetus’ heartbeats. Thereby, we show that a procedure for automatic classification in independent subspaces of the extracted FECG and MECG components is feasible when using a criterion based on the cyclic coherence (CC) of the signal of interest.