Assessing alertness from EEG power spectral bands

Abstract

The assessment of low level of alertness and drowsiness conditions of humans, while performing critical task, requires the development of automatic detection systems to work in real time, to be as pervasive as possible for long lasting periods of use and robust enough to cope whit a wide intra- and inter-individual variability. A new alertness detection procedure based on the spectral analysis of the EEG signal is proposed, mostly concerned with the provision of robust classification criteria under the working conditions depicted above. The wide inter-individual variability has been reduced down to operational levels by means of a personal dependant normalization algorithm, which consists of describing the EEG spectral morphology as a fuction of the alpha behaviour of each subject. With this approach, drownsiness classification can be achieved by simple thresholding of the EEG spectral variable selected: the power ratio between a high frequency and an alpha bands defined for each individual. Variable that has been and its inter-individual stability. The experimental results include the selection of the preferred recording sites and the demostration of the reliability of the classification criteria along the time for each individual. The paper also analyses the time resolution of the algorithms to assure their real time operation. Technological requirements of the method proposed allow concluding that the desing of a wearable one single EEG lead nonintrusive device it is feasible to reliably discriminate continuously drowsiness situations.