Blind separation of ballistocardiogram from EEG via short-and-long-term linear prediction filtering

In this paper the problem of removing ballistocardiogram (BCG) artifact from EEG signals is addressed. This kind of artifact appears in simultaneous EEG-fMRI recordings. We propose a new Blind source extraction method based on linear prediction technique. The proposed method is a joint short-and-long-term prediction (SLTP) strategy to extract the BCG sources. The main reason of using this technique is to jointly model the temporal structure (short-term prediction) of sources and exploiting the prior information of BCG sources (long-term prediction). The results of extensive experiments on both synthetic and real data confirm the strength of the proposed technique to effectively remove the BCG artifact.