Removal of ballistocardiogram artifact from EEG data acquired in the MRI scanner: Selection of ICA components

The removal of ballistocardiogram (BCG) artifact from the EEG recorded in the MRI scanner is challenging. Few studies have utilized independent component analysis (ICA) in this task. A drawback of ICA has been the proper selection of the BCG related components. The key idea in this work is to use the difference between the power spectrum of the artifact-processed data and the spectrum of data recorded outside the scanner as a cost function in the selection of the BCG related independent components. Forward floating selection algorithm was implemented to find the components to minimize this criterion. Also, the typical component selection criteria based on the correlation with electrocardiogram (ECG) signal and on explained variance were compared in this respect. The correlation criterion was least successful leaving considerable residual artifact in the signal. With the first few removed components the variance criterion performed as well as the minimum spectral difference criterion. With the variance criterion alone, however, the number of the components to be removed cannot be determined. The suggested methods may provide objective means to validate residual artifact or the possible loss of physiological signal due to artifact removal and to help selecting the proper artifact-related components.