Fetal cardiac signal extraction from magnetocardiographic data using a probabilistic algorithm

Abstract

Fetal magnetocardiographic sensor measurements are contaminated by undesired environmental and biological signals, such as the maternal cardiac signal. Several methods have been used in an attempt to extract the fetal cardiac signal from these data, which are based on, e.g., the presumed quasi-periodicity of the maternal cardiac signal or the presumed statistical independence between the fetal cardiac signal and interfering signals. Recently a different type of method for extracting signals from noisy data has been introduced. This probabilistic method, known as partitioned factor analysis (PFA), assumes that the data can be partitioned into periods of source inactivity and source activity. PFA was originally developed for stimulus-evoked, trial-averaged encephalographic data, for which the partitions are known in advance. Here we show how to use PFA for extracting the fetal cardiac signal from cardiographic data, for which the partitions must be determined from the data. In addition, we show that PFA can be used even when the partitions cannot be determined directly from the data.