'Signal subspace' blind source separation applied to fetal magnetocardiographic signals extraction

In this paper we apply Independent Component Analysis to magnetocardiographic data recorded from the abdomen of pregnant women. In particular, we include a dimensionality reduction in the 'Cumulant Based Iterative Inversion' algorithm to achieve a 'signal subspace' subdivision, which enhances the algorithm's efficacy in resolving the signals of interest from the recorded traces. Our results show that the proposed two-step procedure is a powerful means for the extraction of the cardiac signals from the background noise and for a sharp separation of the baby's heart from the mother's.