MED: Muse™-based Eye-blink Detection Algorithm Using a Single EEG Channel

Abstract

Eye-blinks in electroencephalogram (EEG) signals can be regarded either as unwanted noise or as a source of information. In both cases, a reliable and accurate detector is needed. As many applications require detection and processing of eye-blinks in real-time, detectors are required to be fast and simple. In this work, we have developed a non-learning algorithm for the detection and extraction of eye-blink segments from EEG signals. The signals were recorded by Muse™, a portable EEG device for recreational use. The proposed algorithm detects eye-blinks via several deterministic processing steps. The algorithm extracts peaks occurring in the EEG signal during the two main eye-blink phases, via extraction of unique features of the EEG eye-blink signal. The proposed algorithm applies various pre-processing steps to ensure robust detection, as well as several sanity-checks to prevent the detection of false peaks and partial eye-blinks. A dataset with recordings of the length of approximately 20 seconds each, taken from few different subjects has been created. The eye-blink annotations were made manually. The proposed algorithm obtains an accuracy rate of 100% on the obtained dataset, while employing a set of deterministic operations which renders it usable in low-resource, real-time applications.