Quantitative Independent Component Selection Using Attractor Analysis for Noise Reduction in Magnetocardiogram Signals

We studied a method for reduction of environmental magnetic noise in magnetocardiograms using independent component analysis, which is a useful technique to separate signal from noise. However, this separation usually relies on subjective choices of the independent components to reconstruct the signal. Hence, we propose a method for quantitative component selection using attractor analysis and coefficient of multiple determination. We verified the noise reduction accuracy using low signal-to-noise ratio experimental data (0 and -10 dB). First, a threshold derived from the coefficient of multiple determination was determined from the results for independent component selection. Then, noise reduction accuracy was evaluated from the spatial correlation coefficients obtained using the proposed method. The results show a considerable reduction of noise for a threshold of 0.6, with its accuracy being almost equal to the maximum noise reduction accuracy obtained from the best selection pattern.