Abstract
Objective:
Removing ocular artefacts due to eye blinks and movement is an essential preprocessing step in an electroencephalogram (EEG) analysis. Most
ocular correction algorithms are based on linear regression or blind source separation algorithms such as independent component analysis (ICA).
Despite their popularity and wide applicability, they also show several areas for improvement such as requiring separate electrooculogram (EOG)
signals or distortions of the EEG signals due to overcorrections.
Methods:
Preliminary studies have shown tensor decomposition as a promising alternative to the ocular artifact correction problem. To extend this line of
research, we propose the SPECTER algorithm, which is the Signal sPECtrum Tensor decomposition and Eye blink Removal algorithm.
Results:
On real data, the algorithm provided comparable or superior performance to ICA- or regression-based artifact correction
methods and outperformed existing tensor-based approaches for eye blink removal. SPECTER also leads to accurate results when traditional eye blink correction methods distort EEG signal.
Conclusion:
In this study, we do not aim to compete with ICA, regression-based, or other eye blink removal approaches, whose good performance has been
proven in the literature. We propose the SPECTER algorithm as an alternative and flexible method in situations where traditional algorithms may fail or identification of the
latent eye blink tensor components is preferred while inspecting EEG data.
Significance:
SPECTER's functionality extends beyond eye blink removal, allowing it to remove a variety of other artifacts, or even more specific EEG rhythms or other EEG elements.
Go back