Blink and saccade detection from forehead EEG

Ocular artifacts that are commonly found in electroencephalography (EEG) recordings are typically discarded as noise; however, this ocular data has the potential to assist with neuroergonomic assessment of cognitive state. In this paper, we develop and evaluate regression models to extract ocular data (e.g., blink rate, saccade rate) from EEG recordings. We used EEG data recorded from participants performing the Multi-Attribute Task Battery simulation at low, medium, and high workloads, with electrooculography (EOG) data recorded simultaneously as a ground truth. Linear regression models were developed to predict EOG data from forehead EEG channels. EOG data were predicted with correlations between 0.72 and 0.94. Blinks were detected in 45x5 min EEG segments per participant with mean precision and recall of 81% and 79% respectfully. Saccades were detected with mean precision and recall of 75% and 76%, respectfully. Blink rates were identified from EEG at similar rates and relationships across workloads. The saccade rate was more varied depending on the participant and may require further investigation to improve.