EEG–fMRI: Dictionary learning for removal of ballistocardiogram artifact from EEG

Abstract

In this paper, artifact removal from biomedical signals is addressed. We particularly focus on removing ballistocardiogram (BCG) artifact from EEG. BCG mainly appears in EEG signals during simultaneous EEG–fMRI recordings. Different from most existing artifact removal techniques, we propose a method based on dictionary learning framework. Due to strength of sparsifying dictionaries in applications such as image denoising, it is expected to succeed in BCG removal task as well. This is investigated in the proposed approach where a dictionary is learned from original EEG recording. The dictionary is designed to locally model BCG characteristics. After achieving the dictionary, BCG can be simply subtracted from the original signal and the clean EEG is obtained. Our experimental results on both synthetic and real data confirm the effectiveness of the proposed method. The results reveal the flexibility of learned dictionary for modeling the fluctuations in artifact, and removing it from original EEG signals.