Performance analysis of blind source separation algorithms in de-noising multichannel magnetocardiogram during rest and stress conditions

Abstract

Magnetocardiography (MCG) offers a non-contact, non-invasive way of capturing the heart's weak magnetic field signals. MCG measurements often suffer from physiological artifacts, particularly during exercise. Blind source separation algorithms (BSS) are commonly utilized to eliminate noise in multichannel systems. This study assesses the performance of BSS algorithms in noise removal from multichannel MCG recordings under varying physical conditions. We used three BSS algorithms such as fast independent component analysis (Fast ICA), temporally decorrelated source separation (TDSEP), and second order blind identification (SOBI) for noise reduction in different physical conditions of the subject rest, stress, and recovery. It has been observed that Fast ICA demonstrated superior performance in rest conditions, while SOBI exhibited better performance during stress conditions, as evidenced by higher signal-to-noise ratio (SNR) across all cardiac beats. This study helps to explore the possibility of improving the quality of MCG signals, facilitating the cardiac health assessment under different physical conditions, especially for subjects with diverse cardiac abnormalities.