Model-Based Filtering of EEG Alpha Waves for Enhanced Accuracy in Dynamic Conditions and Artifact Detection

Abstract

Electroencephalography (EEG) is the recording of brain electrophysiological activity, usually by electrodes placed on the scalp. The EEG signals contain useful information about the brain state, with specific states being associated with oscillations at specific frequencies (the so-called brain waves); hence, EEG signals are usually analyzed in terms of their frequency content. A notable example is the amplitude estimation of alpha waves (8-14 Hz). This paper proposes a model-based estimation approach, based on known physical properties of alpha waves, which allows enhanced robustness in presence of fast amplitude dynamics, as well as an automatic identification of possible artifacts or discontinuities in the alpha wave. The proposed method is illustrated in this paper with application to a clinical EEG signal, but it is particularly promising for wearable EEG applications, such as brain-computer interface (BCI), to name one, where no expert human supervision is available.