Automatic EEG Artifact Removal Techniques by Detecting Influential Independent Components

Abstract

Electroencephalography (EEG) data are used to design useful indicators that act as proxies for detecting humans’ mental activities. However, these electrical signals are susceptible to different forms of interferences—known as artifacts—from voluntarily and involuntarily muscle movements that greatly obscure the information in the signal. It is pertinent to design effective artifact removal techniques (ARTs) capable of removing or reducing the impact of these artifacts. However, most ARTs have been focusing on handling a few specific types, or a single type, of EEG artifacts. EEG processing that generalizes to multiple types of artifacts remains a major challenge. In this paper, we investigate a variety of eight different and typical artifacts that occur in practice. We characterize the spatiotemporal-frequency influence of these EEG artifacts and offer two heuristics. The proposed heuristics extend influential independent component analysis to clean the contaminated EEG signal. These proposed heuristics are compared against four state-of-the-art EEG ARTs using both real and synthesized EEG, collected in the presence of multiple artifacts. The results show that both heuristics offer superior spatiotemporal-frequency performance in automatic artifacts removal and are able to reconstruct clean EEG signals.