To: Introduction | Methods and Materials | Results | Discussion | Conclusions | Acknowledgements | References

Introduction

Human subjects' ability to sustain their initial level of performance during continuous auditory or visual monitoring tasks is limited. After only a few minutes on task, particularly in low-arousal environments, performance of subjects in auditory or visual monitoring tasks often includes periods of intermittent failures to respond to above-threshold targets, alternating with periods of consistent responding. Cycles of relatively good and poor performance tend to last four minutes or longer[4, 18] and are experienced by subjects as alternating waves of drowsiness and alertness[23]. Under sustained low-arousal conditions or when subjects are fatigued, these drowsy periods may progress into sleep episodes. Minute-scale changes in the probability of detection in above-threshold signals during auditory vigilance tasks are accompanied by simultaneous shifts in spectral amplitude of several relatively narrow EEG frequency bands[18], these changes appearing to index the action of brain mechanisms involved in modulating arousal and/or auditory information processing in and near to sleep[26].

One hallmark of the transition from drowsiness to sleep is the appearance of EEG sleep spindles, intermittent oscillatory thalamocortical bursts near 14 Hz[7, 25]. Some early EEG studies of drowsiness also associated lapses in auditory detection with increased EEG amplitude in the theta range (4-7 Hz)[2, 6, 23]. In this study, we explore whether lapses in auditory detection during periods of intermittent detection performance are accompanied by changes in the amplitude of spontaneous EEG activity, and, in particular, whether individual failures to detect auditory target stimuli are more closely associated with increased activity at the sleep spindling frequency or with theta bursts. Finally, we report that different transient perturbations in the EEG spectrum follow presentation of detected and undetected auditory targets and non-target tones.

Return to Summary.

Materials and Methods

Concurrent EEG and behavioral data were collected for the purpose of developing an objective method of monitoring the alertness of operators of complex systems[13, 17, 18]. Fifteen young adult volunteers participated in five half-hour sessions each, during which they pushed one button whenever they detected an above-threshold auditory target stimulus (a brief increase in the level of continuous background noise), and another button whenever they detected a visual pattern on a computer screen in front of them (a slowly-descending line of dots in visual noise).

Stimuli

Auditory targets consisted of 300 ms increases in the intensity of a 62 dB white noise background (150 ms rise, 110 ms fall), presented 6 dB above their threshold of detectibility at pseudorandom time intervals producing a mean rate of 10/min. Brief (50 ms) non-target probe tones of two frequencies (20% 568 Hz, 80% 1098 Hz) were interspersed between the target noise bursts at 2-4 s intervals. Visual targets were vertical lines of dots presented in a slowly descending (2 lines/sec) visual noise 'waterfall' pattern at a mean rate of 1/min. Sessions were conducted in a small, warm, and dimly-lit experimental chamber. Visual task data will be reported elsewhere.

Data Collection

EEG was collected from two electrodes at the vertex and posterior midline (Cz and Pz/Oz, midway between Pz and Oz), referred to the right mastoid, at a sampling rate of 312.5 Hz after analogue filtering (Grass Model 12, bandpass 0.1-100 Hz, 12 dB/octave). Channel locations were selected as likely to contain independent alertness information on the basis of previous studies[18]. Bipolar vertical and horizontal electrooculogram (EOG) data were also recorded for use in artifact rejection. Data windows containing potential deviations from baseline of more than 75 uV were rejected from analysis. Three sessions each from ten subjects containing sufficient (mean +/- s.d., 56+/-26) response lapses were used in the analysis.

Local Error Rate

A time-varying, causal behavioral index of alertness, local error rate, was defined by computing a moving average probability of detection of auditory targets smoothed using a 95-s wide exponential moving window (90% down 95 s back) whose width and shape were chosen to increase smoothness of the resulting measure while decreasing the (23 s) time lag from the leading edge of the window to its center of gravity. Time-varying EEG spectra with 0.6 Hz resolution were computed at 1.64-s intervals using running median averages of FFT transforms of Hanning-windowed 0.82-s data epochs with 50% overlap (zero-padded to 512 points / 1.64 s). The median filtering window spanned seven overlapping 0.82-s data windows, encompassing 3.28 s of data. The four median-filtered spectral estimates in each 1.64-s interval were then averaged.

Error-Sorted Spectra

Changes in the EEG spectrum as a function of error rate were computed by separately averaging log spectra in 4-second epochs preceding detected (hit) and undetected (lapse) target presentations, sorting these spectra according to the prevailing error rate, and then smoothing the results using a Hanning window (encompassing a 30% error-rate range) which was moved through the error-sorted data in 5% error-rate steps. The mean log spectrum during the first 2 min of error-free performance was then subtracted from each spectral trace, yielding hit- and lapse-related error-sorted spectra revealing mean changes in the EEG spectrum accompanying different levels of performance (see Fig. 1).

Spectral Perturbations

The time course of event-related changes in the amplitude of oscillatory EEG activity time locked, but not necessarily phase locked, to experimental events can be measured using normalized time-frequency transforms, previously called event-related spectral perturbations (ERSPs)[16]. Averages of time-frequency transforms of EEG epochs time locked to experimental events can measure event-related modulations of EEG amplitude at a wide range of frequencies and time scales (tenths of seconds[12] to minutes or longer[9, 18]). Time- and frequency-specific ERSP features are largely independent of concurrent features of time-domain response averages (event-related potentials or ERPs). ERPs contain potential deviations occurring at fixed times and in fixed phases relative to experimental events, while ERSPs record changes in the amplitude of EEG oscillations irrespective of their phase at event onsets. Here, ERSPs time locked to hit and lapse target noise bursts, and to non-target tones, were computed by averaging logarithmic time-frequency amplitude transforms of 54-s spectral epochs surrounding stimulus onsets. Two types of normalization were applied (see Fig. 2).

Return to Summary.

Results

Fig. 1a contains two prominent maxima -- relative increases in EEG amplitude near 5 Hz and 14 Hz, plus a smaller increase at frequencies below 5 Hz. At high error rates, mean amplitude near 10 Hz at site Pz/Oz also decreased in most subjects below its eyes-open baseline level. The 14-Hz peak appears only when error rate is above 80%, consistent with the emergence of 14-Hz spindles at sleep onset[7]). At high error rates, the error-sorted spectrum at Cz prior to hits

Return to Summary.

Discussion

During sleep, behavioral and brain responsiveness to sounds is decreased, and communication of afferent auditory information between the thalamus and auditory cortex is restricted or gated[25, 26]. Although a tonic increase in activity at the human sleep spindle frequency (near 14 Hz)[7] accompanies sustained decrements in detection probability, 14-Hz amplitude is not related to the detection of individual targets Fig. 1c). Our analyses do not determine what portion of the 14-Hz peaks in Fig. 1a and 1b reflect the appearance of classic sleep spindles. However, the lack of a 14-Hz feature in Fig. 1c strongly suggests that sleep spindles, hallmarks of transition to slow-wave sleep, cannot be directly involved in intermittent gating of sensory information during drowsiness, although they may have a role in producing tonic changes in auditory sensitivity during sleep onset[25].

The most important conclusion to be drawn from the error-sorted spectra (Fig. 1) is that during periods of intermittent performance, individual detection lapses are associated with increased activity near 5 Hz, not with 14-Hz sleep spindles or increased delta activity. ERSP (Fig. 2) and event-related error-rate (Fig. 3) analyses demonstrate that the observed 4-6 Hz differences arise from phase-locking of 15-20 s EEG and performance cycles.

Phasic Cycles

During periods of inconsistent responding, seven of our ten subjects show 15-20 s cycles in detection probability, closely paralleled by opposing changes in EEG amplitudes near 5 Hz and above 35 Hz. In accord with claims of early EEG studies, undetected targets are preceded by increased activity near 5 Hz, while in our experiments EEG activity above 35 Hz tends to increase prior to detected targets.

Second-order spectral analyses of amplitude variance at each EEG frequency (not shown), applied to the same epochs used to compute the ERSPs (Fig. 2), revealed a peak cycle length between 15 and 20 s per cycle over a wide range of EEG frequencies, consistent with other reports[19]. The 15-20 s theta and gamma perturbations in Fig. 3a, then, may result from mutual phase locking of spontaneous amplitude fluctuations at these frequencies, uncorrelated with 15-20 s fluctuations at lower (< 4 Hz) and intermediate frequencies (10-25 Hz). The fact that the theta-band amplitude cycles are also time locked to performance cycles (Fig. 3a) suggests that intermittent gating of auditory information transmission during drowsiness may be intimately connected to 4-6 Hz activity.

While potential neurophysiological mechanisms for this phenomena have been discussed[25, 26], the behavioral and physiological distinction we report here between spindling-frequency activity and the sleep-related theta rhythm does not seem to have been observed in animal brains. [NOTE: However, 20 s cycles in EEG amplitude have been shown to be synchronized to sleep onset in neocortex of rats (Oakson & Steriade, 1983, Fig. 3]. Note that we cannot rule out possibilities that some or all lapses in these experiments arose from subjects sleep-related inability to recall the significance of the tones, or from an inability to press the response button, although in our own experience as subject (tpj) and pilot subject (sm), all three dimensions of awareness (auditory, situational, and somatic) seemed to covary.

This is not the first report that changes in theta and gamma band EEG amplitudes are linked to changes in attention and awareness[8, 11]. Intermittent vigilance decrements similar to the 15-20 s cycles we report here have recently been reported in monkeys performing a continuous visual task, and have been correlated with cyclic decreases in locus coeruleus activity[21]. Conversely, noradrenergic locus coeruleus activation is known to produce behavioral alerting, accompanied by a rapid shift of the cortical EEG spectrum from low to high frequencies[27]. Possibly, a similar modulator may underlie the 15-20 s performance and EEG fluctuations seen both in quietly resting[19] and here in drowsy humans. More than one brain arousal subsystem may also be involved in producing them[25].

Our observed pre-stimulus increase in theta-band EEG prior to lapses in these experiments confirms and further quantifies early reports[6], most notably a 1962 study of Williams et al.28 in which this phenomenon was seen in EEG tracings of some subjects after 30 hours of sleep deprivation. In other early EEG studies, "low-level irregular" theta activity was associated with self-reports of daydreaming and "floating" sensations typical of the hypnagogic period preceding sleep onset[23]. Most recent studies of theta band EEG, however, have focussed on dissimilar phenomena associated with arousal and attention, including rhythmic slow activity (RSA) in the septohippocampal and related system appearing in animals during volitional movements[14] and frontal midline theta EEG in humans associated with mental effort[10].

The increased gamma-band activity preceding detected targets in our data is compatible with hypotheses that transient phase correlations among local gamma band oscillations in the cortex and thalamus may help support allocation of perceptual attention[5, 8, 15, 22]. Although we cannot rule out possible contributions of muscle activity to changes in gamma band activity prior to detected or undetected targets, muscle activity normally contains frequencies below 35 Hz, the lower edge of our performance-related gamma-band difference, suggesting that the phasic gamma band phenomenon shown in Fig. 2 arises predominately from performance-related changes in brain activity.

Transient Perturbations

Successfully detected auditory targets are followed by a brief increase in 10-25 Hz EEG activity beginning a second or more after stimulus onset, while undetected targets are followed by a brief relative amplitude decrease in the same time and frequency range.

The appearance of 10-25 Hz spectral perturbations following (but not preceding) undetected (Fig. 2b) as well as detected (Fig. 2c) targets implies that the brain system responsible for these EEG modulations also detects lapse targets, to which subjects do not respond. ERSPs (not shown) time locked to onsets of the frequent (16/min) non-target tones in these sessions contained no comparable response feature, implying that the EEG modulator responsible for the 10-25 Hz response can also distinguish target noise bursts from non-target tones, even though the non-target tones in these experiments were perceptually more salient than the relatively faint target noise bursts. Similarly, in auditory experiments reported earlier[16], a ridge of augmented 11-19 Hz activity induced by very brief target tone pips was found to be twice as large that induced by intense (84 dB SL), long-lasting (1 s) non-target tones.

Note that this attention-related 10-25 Hz perturbation appears much later than attention-related features of the auditory ERP, such as the P300, but closer in latency to attention-related autonomic responses such as the evoked pupillary response1. In general, the magnitudes of ERSP features (0.5 - 2 dB, 5 - 25%) in this and previously-reported experiments are comparable in size to task-related differences in local brain metabolism reported in many PET and fMRI brain studies[20, 24], suggesting a possible relation between EEG spectral perturbations and the size of local brain blood-flow changes.

Return to Summary.

Conclusions

Concurrent activity on a wide range of time scales is a basic fact of brain physiology[3]. Our results show that during periods of impaired performance on an auditory vigilance task, at least three time scales characterize the dynamics of variations in auditory detection performance and the EEG spectrum. Overall, changes in error rate in auditory detection are dominated by irregular waves of performance changes lasting four minutes and longer[4, 18]. Within these, phasic 15-20 s cycles in detection performance are phase locked to counterbalanced changes in EEG amplitudes near 5 Hz and above 35 Hz, respectively. Finally, auditory stimuli induce transient and variable perturbations in the EEG amplitude spectrum which can be observed by averaging normalized time-frequency transforms (ERSPs) of event-related data epochs. In particular, faint target noise bursts (but not more salient non-target tones) in our experiments induce transient time locked perturbations in mean EEG amplitudes at intermediate frequencies (10-25 Hz), whose sign (positive or negative) depends on whether or not the subject responds to the target. Further understanding of characteristic time and frequency scales involved in cognitive brain processing during sleep and waking may lead to a deeper appreciation of the organization of human consciousness and performance, and may also have practical applications in monitoring the alertness of operators of complex systems[13, 17].

Acknowledgements

Return to Summary.

1. Beatty, J., Phasic not tonic pupillary responses vary with auditory vigilance performance, Psychophysiology, 19 (1982) 167-172.

2. Beatty, J., Greenberg, A., Deibler, W.P., and O'Hanlon J.F., Operant control of occipital theta rhythm affects performance in a radar monitoring task. Science, 183 (1974) 871-873.

3. Churchland, P.S, .and Sejnowski, T.J., Perspectives on cognitive neuroscience. Science, 242 (1988) 741-745.

4. Conte, S., Ferlazzo, F., and Renzi, P., Ultradian rhythms of reaction times in performance in vigilance tasks. Biolog. Psychol., 399 (1995) 159-172.

5. Crick. F., and Koch, C., Some reflections on visual awareness. Cold Spring Harbor Sympos. on Quant. Biol., 55 (1990) 953-962.

6. Davis, H., Davis, P.A., Loomis, A.L., Harvey, E.N., and Hobart, G., Human brain potentials during the onset of sleep. J. Neurophysiol., 2 (1937) 24-37.

7. Dijk, D-J., Hayes, B., and Czeisler, C.A., Dynamics of electroencephalographic sleep spindles and slow wave activity in men: effect of sleep deprivation. Brain Res., 626 (1993) 190-199.

8. Freeman, W., Mass Action in the Nervous System, Academic Press, New York, 1975, 489 pp.

9. Galambos, R., and Makeig, S., Dynamic changes in steady-state responses. In: H. Basar (Ed.), Dynamics of Sensory and Cognitive Processing by the Brain, Springer-Verlag, Berlin, 1988, pp. 103-122.

10. Inouye, T., Shimosako, K., Iyama, A., Matsumoto, Y., and Toi, S., Moving potential field of frontal midline theta activity during a mental task. Cog. Brain Res., 2 (1994) 87-92.

11. Jahnsen, H., and Llinas, R., Electrophysiological properties of guinea-pig thalamic neurons: an in vitro study. J. Physiol. Lond., 349 (1984) 205-225.

12. Jokeit, H., and Makeig, S., Different event-related patterns of gamma-band power in brain waves of fast- and slow-reacting subjects. Proc. Nat. Acad. Sci. USA, 91 (1994) 6339-6343.

13. Jung, T-P., and Makeig, S., Predicting failures in auditory detection from changes in the EEG spectrum, Abstract, IEEE Soc. for Engin. in Med. and Biol. (1995).

14. Leung, L-W.S., Lopes da Silva, F.H., and Wadman, W.J., Spectral characteristics of the hippocampal EEG in the freely moving rat. Electroencephalog. clin. Neurophysiolog., 54 (1982) 203-219.

15. Lisman, J.E., and Idiart, M.A.P., Storage of 72 short-term memories in oscillatory subcycles. Science, 267 (1995) 1512-1516.

16. Makeig, S., Auditory event-related dynamics of the EEG spectrum and effects of exposure to tones. Electroencephalog. clin. Neurophysiolog., 86 (1993) 283-293.

17. Makeig, S. Elliott, F.S., and Postal M. First Demonstration of an Alertness Monitoring and Management System. Naval Health Research Center, San Diego, CA, 1993, Tech. Report 93-36.

18. Makeig, S. and Inlow, M., Lapses in alertness: coherence of fluctuations in performance and EEG spectrum. Electroencephalog. clin. Neurophysiolog., 86 (1993) 23-35.

19. Novak, P., Lepicovska, V., and Dostalek, C., Periodic amplitude modulation of EEG. Neuroscience Lett. 136 (1992) 213-215.

20. Petersen, S.E., Fox, P.T., Snyder, A.Z., and Raichle, M.E., Activation of extrastriate and frontal cortical areas by visual words and word-like stimuli. Science, 249 (1990) 1041-1044.

21. Rajkowski, J., Kubiak, P., and Aston-Jones, G., Locus coeruleus activity in monkey: phasic and tonic changes are associated with altered vigilance. Brain Res. Bull., 35 (1994) 607-616.

22. Rogeul, A., Bouyer, J.J., Dedet, L., and Debray, O. Fast somato-parietal rhythms during combined focal attention and immobility in baboon and squirrel monkey. Electroencephalog. clin. Neurophysiolog., 46 (1979) 310-319.

23. Schacter, D.L. The hypnagogic state: a critical review of the literature. Psycholog. Bull., 83 (1976) 452-481.

24. Schneider, W., Noll, D.C., and Cohen, J.D., Functional topographic mapping of the cortical ribbon in human vision with conventional MRI scanners. Nature, 365 (1993) 150-153.

25. Steriade, M., Jones, E.G. and Llinas, R.R. Thalamic Oscillations and Signaling, Wiley, New York, 1990, 431 pp.

26. Steriade, M., McCormick,D.A., and Sejnowski, T.J., Thalamocortical oscillations in the sleeping and aroused brain. Science, 262 (1993) 679-688.

27. Swick, D., Pineda, J.A., Schacher, S., and Foote, S.L., Locus coeruleus neuronal activity in awake monkeys: relationship to auditory P300-like potentials and spontaneous EEG. Exp. Brain Res., 101 (1994) 86-92.

28. Williams, H.A., Granda, A., Jones, R., Lubin, A., and Armington, J., EEG frequency and finger pulse volume as predictors of reaction time during sleep loss. Electroencephalog. clin. Neurophysiolog., 14 (1962) 64-70.

Return to Summary.