Blind source separation of fetal ECG using fast independent component analysis and particle swarm optimization

Abstract

One of the main challenges in prenatal monitoring is the extraction of non-invasive fetal electrocardiogram (fECG) from maternal abdominal recordings. This thesis compares two different signal processing methods for fECG extraction, Fast Independent Component Analysis (FastICA) and Particle Swarm Optimization (PSO). Both methods are applied to multichannel abdominal recordings of the ADFECGDB (PhysioNet). The performance is measured by SNR, MSE, correlation, true positive rate. Results show that PSO substantially outperforms FastICA. PSO-based approach is more robust than FastICA for fetal ECG extraction in real-world