Robust 3-way tensor decomposition and extended state Kalman filtering to extract fetal ECG

This paper addresses the problem of fetal electrocardiogram (ECG) extraction from multichannel recordings. The proposed two-step method, which is applicable to as few as two channels, relies on (i) a deterministic tensor decomposition approach, (ii) a Kalman filtering. Tensor decomposition criteria that are robust to outliers are proposed and used to better track weak traces of the fetal ECG. Then, the state parameters used within an extended realistic nonlinear dynamic model for extraction of N ECGs from M mixtures of several ECGs and noise are estimated from the loading matrices provided by the first step. Application of the proposed method on actual data shows its significantly superior performance in comparison to the classic methods.