EEG mobility artifact removal for ambulatory epileptic seizure prediction applications

Abstract

Mobile monitoring of electroencephalogram (EEG) signals is prone to different sources of artifacts. Most importantly, motion-related artifacts present a major challenge hindering the clean acquisition of EEG data as they spread all over the scalp and across all frequency bands. This leads to additional complexity in the development of neurologically-oriented mobile health solutions. Among the top five most common neurological disorders, epilepsy has increasingly relied on EEG for diagnosis. Separate methods have been used to classify EEG segments in the context of epilepsy while reducing the existing mobility artifacts. This work specifically devises an approach to remove motion-related artifacts in the context of epilepsy. The proposed approach first includes the recording of EEG signals using a wearable EEG headset. The recorded signals are then colored by some motion artifacts generated in a lab-controlled experiment. This stage is followed by temporal and spectral characterization of the signals and artifact removal using independent component analysis (ICA). The proposed approach is tested using real clinical EEG data and results showed an average increase in accuracy of ∼9% in seizure detection and ∼24% in prediction. © 2019 Elsevier Ltd