Regression Approach for Automatic Detection of Attention Lapses

Abstract

Certain professions rely on the ability to maintain attention constant throughout long periods of time, like truck drivers, air traffic controllers, health professionals, among others. These could greatly benefit from the development of a real-time alerting system that will call subjects back to task even before lapses occur or shortly after they happened. Attention levels have been shown to relate to the properties of the electroencephalogram (EEG). In this paper, we propose for the first time a regression approach to detect fluctuating levels of attention, based on spatiotemporal patterns extracted from EEG recordings. Previous studies have shown that reaction time is related to the level of task related attention. Moment-to-moment fluctuations in attention level are paralleled by moment-tomoment fluctuations in reaction time (faster reaction times are related to high attention allocation). We took advantage of this parallel and used reaction time data obtained during a repetitive visuomotor task as a proxy for task related attention level. Furthermore, instead of defining high attention versus low attention periods, we labeled each moment according to a continuum based on each trial's reaction time. In order to determine if it is possible to predict attention level from EEG features, we developed regression models between the extracted features and the subject's reaction time.