Blind extraction of fetal and maternal components from the abdominal electrocardiogram: An ICA implementation for low-dimensional recordings

Abstract

This work studied Space-Time ICA (ST-ICA) on the separation of fetal and maternal ECGs from low-dimensional abdominal recordings. ST-ICA was implemented by combining the Method of Delays with FastICA and an automatic classifier and applied to the “Abdominal and Direct Fetal Electrocardiogram Database” in Physionet. Next, 24 indexes qualified the FECGs retrieved by using 4, 3, 2 channels and combinations of them (e.g., channels 1–3–4 or channels 2–4) with different configurations of FastICA (e.g., deflation and pow3). Results in five subjects indicate that, while the traditional implementation of FastICA only separated low quality FECGs in two subjects, ST-FastICA is more likely to consistently separate high quality FECGs in all subjects (even though the fetal and maternal QRS complexes were overlapping), this by using four channels, in the symmetric mode, independently on the nonlinearity. Results when using 3 and 2 channels also indicated that ST-FastICA is suitable for the separation of fetal and maternal ECGs, although the quality of the FECG extracted depended on the combination given by the channels used and the configurations of FastICA. Finally, when calculating RR intervals and fetal heart rate, ST-ICA (with 4 channels) showed no statistical differences with the Varanini et al. algorithm [1], one of the top-performing entries in the CinC13 Challenge. These preliminary results indicate that the temporal dynamics and the spatial information used by ST-FastICA make it possible for it to achieve a robust separation of ECGs in low-dimensional recordings where the traditional implementation of FastICA is more likely to fail.