Fetal ECG extraction from sparse representation of multichannel abdominal recordings

Noninvasive extraction of fetal electrocardiogram (ECG) from maternal abdominal recordings is quite challenging as such signals are often corrupted by signals from other sources, with maternal heart activities being the most distorting one. In this paper, a modified compressive sensing (CS)-based approach for extracting fetal ECG signals from multichannel abdominal recordings is proposed. Sparse representations of the acquired abdominal recordings allows for the effective compression rate of 75% for the recorded data. The scheme deploys two BSS algorithms, namely fast independent component analysis (fICA) and time–frequency BSS (TF–BSS), to estimate the source signals from the recordings and extract the fetal ECG. The performance of the proposed method is evaluated using the publicly available 2013 PhysioNet Challenge database and compared with that of the best performing existing ones. The experimental results show that the proposed framework outperforms the existing methods with a mean minimum square error of 98.59 and exhibits computational complexity comparable with the best existing methods. The results also show that the discrete wavelet transform dictionary performs well as sparsifying basis for abdominal recordings in a CS-based fetal ECG extraction framework. The proposed method can therefore be used for noninvasive and reliable extraction of fetal ECG from abdominal recordings and for developing wireless body sensor networks for ECG tele-monitoring.