Linear and nonlinear features for automatic artifacts removal from MEG data based on ICA

This paper presents an automatic method to remove physiological artifacts from magnetoencephalogram (MEG) data based on independent component analysis (ICA). The proposed features including kurtosis (K), probability density (PD), central moment of frequency (CMoF), spectral entropy (SpecEn), and fractal dimension (FD) were used to identify the artifactual components such as cardiac, ocular, muscular, and sudden high-amplitude changes. For an ocular artifact, the frontal head region (FHR) thresholding was proposed. In this paper, ICA method was on the basis of FastICA algorithm to decompose the underlying sources in MEG data. Then, the corresponding ICs responsible for artifacts were identified by means of appropriate parameters. Comparison between MEG and artifactual components showed the statistical significant results at p <; 0.001 for all features. The output artifact-free MEG waveforms showed the applicability of the proposed method in removing artifactual components.