Eye blink characterization from frontal EEG electrodes using source separation and pattern recognition algorithms

Due to its major safety applications, including safe driving, mental fatigue estimation is a rapidly growing research topic in the engineering field. Most current mental fatigue monitoring systems analyze brain activity through electro-encephalography (EEG). Yet eye blink analysis can also be added to help characterize fatigue states. It usually requires the use of additional devices, such as EOG electrodes, uncomfortable to wear, or more expensive eye trackers. However, in this article, a method is proposed to evaluate eye blink parameters using frontal EEG electrodes only. EEG signals, which are generally corrupted by ocular artifacts, are decomposed into sources by means of a source separation algorithm. Sources are then automatically classified into ocular or non-ocular sources using temporal, spatial and frequency features. The selected ocular source is back propagated in the signal space and used to localize blinks by means of an adaptive threshold, and then to characterize detected blinks. The method, validated on 11 different subjects, does not require any prior tuning when applied to a new subject, which makes it subject-independent. The vertical EOG signal was recorded during an experiment lasting 90 min in which the participants’ mental fatigue increased. The blinks extracted from this signal were compared to those extracted using frontal EEG electrodes. Very good performances were obtained with a true detection rate of 89% and a false alarm rate of 3%. The correlation between the blink parameters extracted from both recording modalities was 0.81 in average.