Ocular Artifacts Elimination From Multivariate EEG Signal Using Frequency-Spatial Filtering

The electroencephalogram (EEG) signals record electrical activities generated by the brain cells and are used as a state-of-the-art diagnosis tool for various neural disorders. However, the unwanted artifacts often contaminate the recorded EEG signals and disturb the interpretation of the neuronal activity. This article aims to propose an efficient automatic method to eliminate the ocular artifacts (OAs) from the multichannel EEG signals with novel frequency-spatial filtering. The method combines dictionary-based spatial filtering and frequency-based signal decomposition method, namely, empirical wavelet transform (EWT). The artifact dictionary needed for spatial filtering is isolated from the raw data by: 1) selecting the contaminated channels and 2) frequency-domain filtering. More precisely, the
$\delta $
-rhythms of identified highly contaminated channels are selected and placed into an artifact dictionary. Afterward, the
$\delta $
-rhythms of multichannel EEG signals are spatially filtered using the built dictionary to seclude the OAs within a limited number of components. Furthermore, the artifact components are eliminated and clean
$\delta $
-rhythms are recovered using the inverse spatial filtering technique. Finally, the clean
$\delta $
-rhythms are combined with other EEG rhythms to reconstruct the OA-free signals. The proposed method is applied to OA-contaminated synthetic and real multichannel EEG signals with a convincing performance as compared to state-of-the-art approaches. The proposed method removes the OAs without affecting the background EEG information. The proposed method can ease sensor signal interpretation and further processing, e.g., for BCI applications.