Variance-based spatial filtering in fMCG

The objective is to separate fetal heart signals and maternal heart interference in the fetal magnetocardiogram (fMCG), In a multi-channel (37-channel) detector, the measurement encodes the spatial information of signals. In our previous work, we used the spatial filtering based on eigen-vector constraints to separate fetal and maternal signals. That approach was very successful in most of the cases, however, it had the problem of signal distortion. In a few cases, the noise was amplified, In this paper, we attempt to deal with the noise problem by using variance-based spatial filtering, A time sample defines a vector in the 37-dimensional ambient space, For each type of signal, we used a sequence of vectors from a single or several waveforms that correspond to the signal to depict its spatial information. We then applied principal component analysis on those sequences of vectors for each signal to extract their principal directions (spatial patterns), The results are used in the linear constrained minimum variance (LCMV) method for both the forward solution matrix and the covariance matrix. Applications of this method on some real data show improved SNR, But, in general, it does not separate signals as thoroughly as the direct transforms. This method allows flexible choices for the forward matrix and the covariance matrix, which can be used to preserve multiple types of signals.