Introduction

Patterns of electrical activity of the brain of healthy human subjects (electroencephalogram, EEG) depend on a variety of factors, including various aspects of behaviour. These include the basic EEG type of the subject, level of alertness and the type of mental activity being carried out. The EEG, recorded at multiple sites over the scalp, also shows spatial variations, with different brain areas co-operating during mental activity. These relationships between alertness, mental activity and the EEG are examples of the links between EEG activity and behaviour. The purpose of the present study was to use the EEG to predict errors in performance brought about by fatigue during an overnight period of work. During this study, behaviour was manipulated, firstly, by inducing fatigue, and secondly, by requiring the subjects to perform different types of mental activity. The mental tasks involved activities which are included in the performance of complex tasks, namely memory, attention, co-ordination, and mathematical and spatial processing ability.

EEG recordings and cognitive tasks

Brain activity was recorded from 19 electrode sites over the scalp, using an elasticated electrode cap (ECI Electro-Cap) while subjects carried out a variety of cognitive performance tasks. The EEG signals were recorded by a polygraph machine (Medelec 1121 series) and digitised using a commercially available data acquisition system (DATS, Prosig Computer Consultants Ltd.). Tests of memory function, vigilance, target detection, psychomotor performance (target tracking), mathematical processing and visual rotation were included, and the spontaneous EEG during rest was also recorded. The experiment consisted of an overnight period of work lasting approximately 18 hours, including the period between 23:00 and 06:30, and was intended to produce fatigue. The tasks were each carried out on seven occasions, in sessions lasting approximately 70 minutes. Six healthy subjects participated in the study, and were trained to plateau level before commencing.

Analysis procedure

The EEG was analysed using the Fast Fourier transform, based on epochs of data lasting 2 seconds, which were then meaned over several minutes (typically three to seven minutes corresponding to the duration of the task) to obtain stable estimates of power spectra. Regression analyses were then used to predict error rates in task performance from the EEG. The EEG parameters used in the regression were based on groups of electrode sites determined using analysis of variance and multiple comparison procedures, and on variables derived from the EEG spectra using principal components analysis. The EEG variables covered the delta, theta, alpha and beta frequencies of the spectrum. These frequencies are known to alter during task performance and when alertness decreases.

Results

During the later sessions of the experiment, task performance was reduced to levels of 30 to 40% of those seen during the daytime when subjects were alert and well-rested. The EEG data showed high correlations with performance, based on the relationships between groups of EEG channels and the percentage of correct responses for individual tasks. The results indicated that the EEG can predict decrements in cognitive performance in terms of the percentage of correct responses to within 10 to 20% accuracy. These predictions were based on the average value of power from groups of EEG sites, involving areas of the brain concerned with intellectual function (frontal and pre-frontal sites), memory (temporal sites) and motor function (central sites), and also regions which normally show changes as alertness decreases (parietal and occipital sites).

Discussion

The present study has demonstrated the capability of brain activity to predict decrements in cognitive performance, including memory, vigilance and target tracking. As such, they provide a clear link between the EEG and behaviour, specifically performance at a variety of cognitive tasks, which has a practical application. The predictions were from data involving all subjects and tasks combined, using a model that allowed for sources of variation due to subjects and tasks. Given that individual variability exists between the subjects’ EEGs and also between tasks, accuracy of the predictions was acceptable in the majority of cases with this group of subjects for the tasks in the schedule. These predictions, using EEG recordings from 19 sites on the scalp, improve upon those based on EEG activity at a single electrode, using individual subjects’ EEGs [1], where only extreme degradations in performance (less than 25% correct) were able to be detected by the EEG. The findings were, however, based on relatively long segments of EEG data covering the duration of various performance tasks and typically lasting several minutes. The overall aim of the current research is to predict performance in a real-time situation in occupational settings so that loss of alertness and errors in performance may be identified and prevented. Clearly, this will need to be carried out in a dynamic manner on a considerably shorter time scale, of the order of a few seconds. Approaches to the above issues will be discussed.

References

1. Belyavin, A.; Wright, N.A. (1987). Changes in electrical activity of the brain with vigilance. Electroenceph. and Clin. Neurophysiol., 66, 137-144.