Attention State Classification with In-Ear EEG

Abstract

Electroencephalography entirely inside the ear (in-ear EEG) opens up exciting avenues for unobtrusive continuous physiological and cognitive state monitoring. This work presents techniques towards precise attention state classification based on data recorded from comfortable, binaural in-ear EEG instruments used during a vigilance task experiment. We recorded both on-scalp and in-ear EEG signals from multiple subjects, and we show that in-ear EEG offers comparable classification accuracy. Our work is the first adaptation of Common-mode spatial filtering techniques applied to signals acquired from sparse electrodes on untethered subjects. We demonstrate 90-95% accuracy (scalp-EEG with 30 electrodes) and 70-75% (in-ear EEG with 5 electrodes inside the ear canal and concha) in classifying attentive and resting states. We also show our approach to be lightweight for low-power on-chip classification with the capacity for few-shot learning. A necessity for wearable, continuous health sensors accommodating resource-constrained applications and adaptation to inter-subject differences and varying environmental conditions. This study suggests the future viability for a System-on-Chip (SoC) integration for user-generic and portable devices capable of closed-loop cognitive state monitoring and neuro-feedback.