2.3 The EEG and Detection of Drowsiness

The EEG is a common physiological indicator used to measure wakefulness and sleepiness.  There is considerable evidence that physiological sleepiness is directly related to the rapidity of onset of EEG-defined sleep.  For a relaxed subject with their eyes closed, sleep onset generally involves the substitution of theta activity for the alpha activity that is normally present in the relaxed wakeful state.  It has been observed that the transition to sleep is also frequently accompanied by slow rolling eye movements (SEMs) that can be detected in EOG signals.

Detecting the condition of sleep in a pilot or driver may not always mitigate a potentially dangerous situation; if the individual is already in the sleep state, even if awakened, there may be insufficient time to avoid an impending accident. It is more important to detect the condition of extreme sleepiness, which refers to a state during which the individual struggles against sleep, attention lapses occur, and sleep eventually ensues.  The state of extreme sleepiness is generally manifested by degradation in alertness, vigilance, reactions and reflexes, occurring prior to the onset of sleep.

Research has been conducted in the behavioral assessment of vigilance to evaluate the effects of sleep deprivation on alertness.  These vigilance/alertness experiments generally require that subjects respond to repetitive stimuli, referred to as reaction time tasks [Molodofsky, 1992].  Any delays, lack of response, or errors in the response are examined as potential indicators of sleepiness.  The Wilkinson four choice reaction time method assesses vigilance by determining the detection rate or proportion of correct responses to a series of repetitive stimuli [Wilkinson and Houghton, 1975].

The relationship between performance degradation (slower reaction rates and attention lapses) and increased sleepiness has been established by several researchers [Wilkinson and Houghton, 1975; O'Hanlon and Kelley, 1977; Dinges, 1988; Molodofsky, 1992; Trejo and Shensa, 1993; Makeig and Inlow, 1993].  Similarly, there is also a strong correlation between performance degradation and particular patterns in the EEG waveform [Horvath et al., 1976; O'Hanlon and Beatty, 1977; O'Hanlon and Kelley, 1977; Makeig and Inlow, 1993].  In turn, the EEG waveform has also been correlated with vigilance and sleepiness of subjects in various studies [Gale, 1977; Daniel, 1967; Fruhstorfer, 1977; Santamaria and Chiappa, 1987].

The detection of sleepiness in an individual from the EEG is somewhat different than the detection of sleep.  For instance, whether the eyes are open or closed can make a substantial difference on the level of activity in particular frequency bands of interest in the EEG.  For example as sleepiness develops, a subject whose eyes are closed generally experiences a reduction in the amplitude of the alpha band energy and an increase in the theta band energy present in the occipital channel of an EEG.  However, for an individual with their eyes open, studies clearly indicate that increases in alpha and theta activity in the EEG may reflect sleepiness [Fruhstorfer et al., 1977; O'Hanlon and Kelley, 1977] as well as reduced performance [Daniel 1967; Horvath et al., 1976; O'Hanlon and Beatty, 1977].  Thus, the appearance of alpha activity in the EEG may be an indicator of drowsiness (an incipient indicator of a loss of vigilance), in contrast to the disappearance of alpha activity and a transfer to theta activity in the EEG, which is the case for subjects with their eyes closed [Santamaria and Chiappa, 1987; Makeig and Inlow, 1993].

This somewhat contradictory evidence was explained by O'Hanlon and Beatty [1977], who said that individuals with eyes open show a preponderance of beta (13-30 Hz) activity while alert, which is followed by a shifting of energy to the alpha band as they become drowsy.  The higher frequency beta activity during alert, eyes open periods was often ignored or not evaluated in many of the earlier studies.  Summarizing these conclusions we have that, for individuals with eyes open, drowsiness is associated with an increase in alpha and theta activity and a decrease in beta activity, while for individuals with eyes closed, drowsiness is associated with an increase in theta activity (and possibly delta activity) and a decrease in alpha activity [O'Hanlon and Beatty, 1977; Makeig and Inlow, 1993].  The common conclusion between these results is that drowsiness is associated with a decrease in the frequencies of the predominant energy bands.

In a series of recent papers [Torsvall et al. 1987, 1989; Akerstedt et al., 1990, 1991], several Swedish researchers, led by Torsvall and Akerstedt, studied the effects of drowsiness and sleepiness on shift workers and tried to correlate and identify these effects with the subjects' ambulatory EEG and EOG recordings.  The subjects studied included several train drivers and paper mill workers during the day, afternoon, and night shifts.  A Medilog portable tape recorder was worn by the subjects on a belt and single channel EEG data was recorded using two electrodes, from either O2-P4 or Cz-Oz.  The EOG was also recorded on a single channel for vertical movement or oblique derivation.  The subjects' EKG was also recorded, although it was found that this measurement did not correlate with drowsiness of the subject.

The EOG was scored visually for SEMs, which were defined as slow (< 3/4 Hz) rolling excursions (> 100 V) lasting for more than 1 second.  The EEG signals were sampled at a rate of 68 Hz and were analyzed using spectral analysis methods.  The power density of the signal that was present in particular frequency bands was used to evaluate the EEG.  In addition, the subjects rated themselves at various times on a sleepiness scale with 1 corresponding to "very, very alert", up to 13 corresponding to "very, very sleepy".  Observers were also used in some experiments to record the occurrence of particular signals and events during the shifts.

The results of the studies indicated that rated sleepiness increased sharply during the night shift with similar patterns for SEMs, the power density of the EEG signal in the alpha band, and to a lesser extent, the power density of the EEG signal in the theta and delta bands.  Intra-individual correlations were very high between rated sleepiness, the power density in the alpha and theta bands, and SEMs.  These patterns and correlations were not present during the day and afternoon shifts.

These results were consistent with and substantiated much of the previous research in this area.  Earlier efforts of Torsvall and Akerstedt [1985] also  indicated that "dozing off" is associated with increased power density of the EEG signal in the alpha and theta bands as well as with increased SEM activity.  Figure 2.3.1 contains the EEG spectrum of a night shift train driver displayed as a 3-dimensional graph of power spectrum data obtained using the windowed FFT vs. time.  The increased alpha activity begins after approximately 75 minutes and continues to appear in repeated bursts for the remainder of the shift.

Figure 2.3.1  Windowed FFTs vs. Time (Waterfall Display)
[Torsvall & Akerstedt, 1987]

The two asterisks appearing at approximately 180 minutes and 230 minutes, respectively, represent times when the driver failed to respond to signals.  Both of these events are associated with large alpha activity.  Based on their work, Torsvall and Akerstedt have concluded that, "... it should be possible to use EEG parameters to demonstrate sleepiness at work" [1987].

In another study of subjects that were kept awake and active overnight in a sleep lab, Akerstedt and Gillberg [1990] showed that the intrusion of SEM and alpha and theta power density during waking, open-eyed activity strongly differentiated between high and low subjective sleepiness ratings.  For closed-eyes conditions, it was much more difficult to differentiate between sleepy and alert states (which has also been noticed in sleep stage classification studies [Smith, 1987]).  They therefore concluded that the EEG can sensitively identify states of severe sleepiness given that the subject's eyes are open.  They also noticed a pattern of SEMs during sleepiness with "open eyes" that suggested the subjects alternated between open and almost closed eyes.  This may have been one of the major reasons for the increased alpha activity during drowsiness.

The several experiments of Akerstedt et al. also indicated the importance of the conditions of the subjects and of their surrounding environment.  For instance, with the paper mill workers [Torsvall et al., 1989], the "dozing off" occurred at the time of the trough of the circadian wakefulness rhythm and were associated with low rated work load.  The circadian rhythmicity, sleep loss, and passive work tasks can induce sleepiness during work to the point where wakefulness is difficult to maintain and involuntary sleep ensues.  The train drivers were more likely to inadvertently fall asleep because of their working environment - they had to work alone during darkness, remain seated and keep their attention strictly focused on a very limited area.

Akerstedt et al. [1991] stressed the importance of physical activity during sleepy periods and observed that workers exhibited much lower levels of alpha and theta activity during busier periods.  They suggested that shift workers could prevent much of the manifestation of sleepiness by various types of activity, which may include succumbing to sleep.

Finally, the detection of sleepiness may be more clearly determined from transients in the spectral content of the EEG, rather than from the usual approach of looking at spectral averages over time.  The appearance of SEMs with alpha bursts was characteristic of "dozing off" episodes, which represent a transient failure in fighting off sleep.  In the study of train drivers, large bursts of alpha activity occurred in the most sleepy subjects immediately prior to missed braking signals.  In one case, this pattern continued for about 20 seconds, after which a normal waking EEG pattern appeared as the driver started braking.  Another documented incident in which the driver failed to reduce speed was also accompanied by the continuous appearance of alpha activity.  The paper mill workers that fell asleep while on the job also exhibited very short bursts of alpha and theta activity for several minutes prior to each involuntary sleep episode, as well as during and after it.  In an initial attempt to correlate these bursts with drowsiness, Akerstedt et al. [1991] identified short time segments with alpha and/or theta activity above certain power density thresholds.  The thresholds were defined by the mean power densities at the beginning of the shift (during "alert" conditions).  Theta activity was also above its threshold primarily during sleep, whereas alpha activity was observed to be above its threshold more often prior to the occurrence of sleep [Akerstedt et al., 1991].  Thus, the recent research appears to indicate that alpha activity, theta activity, and SEMs are most sensitive to sleepiness, and that transient changes in alpha and theta activity may be important indicators of drowsiness.

It needs to be pointed out that even though correlations with traditional frequency bands have been made, these correlations have not been consistent enough among individuals, tasks, etc. to be used effectively in a drowsiness detection system.  In fact, current research is focusing on other observables such as eyes (blink rate, duration, etc.), behavior (steering wheel movements, task performance, etc.), and other non-EEG variables.

<PREVIOUS

NEXT>