1Department of Anaesthesia, Waikato Hospital, Hamilton, New Zealand. 2Department of Physics and Medical Electronics, University of Waikato, Hamilton, New Zealand*Corresponding author
Accepted for publication: August 19, 2000
![]() |
Abstract |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Br J Anaesth 2001; 86: 508
Keywords: monitoring, electroencephalography; anaesthesia, general; monitoring, electromyography
![]() |
Introduction |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
However, in recent years the introduction of the bispectral index (BIS) by Aspect Medical Systems (Natick, Massachusetts, USA) has achieved success as an EEG monitor of the conscious state.11 With the publication of some of the algorithms involved in the derivation of the BIS,12 it is apparent that one feature that sets this monitor apart from previous monitors is that it recognizes the importance of the higher EEG frequencies (up to 47 Hz) as indicators of the state of consciousness. The details of the calculation and significance of the components of the BIS are discussed in Appendix 1.
There is a large number of possible alternative estimators that could be used to quantify the higher frequencies. In this paper we take a simple approach and use the first time-derivative of the raw EEG signal. Taking the derivative is mathematically equivalent to scaling the EEG signal by its frequency, thus markedly increasing the relative contribution of the higher frequency components and reducing that of the low-frequency components in a linear fashion. Figure 1 compares the power spectrum of the raw EEG signal (left) against that of the first time-derivative of the EEG signal (right).
|
![]() |
Methods |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
EEG acquisition and analysis
To reduce electrodeskin impedance, the skin over the forehead of each subject was cleaned with an abrasive cleaning fluid (Omniprep; DO Weaver, Aurora, Colorado, USA), and low-impedance electrode paste (Grass EC2 electrode cream; Astro-Med, Warwick, Rhode Island, USA) was placed under disposable adhesive silversilver chloride electrodes (Meditrace 200; Graphic Controls, Buffalo, NY, USA). We have found previously that, using this arrangement, we are able to obtain electrode impedances similar to, or better than, those obtained with the proprietary disposable EEG electrodes. The EEG was monitored using a bipolar bifrontal montage format (Fp1Fp2, 10/20 system). The ground electrode was placed at the mid-forehead (Fz) position. In the volunteer group, a second bipolar EEG channel was recorded from the temporoparietal region (T3P3). In a subset of 30 of the patient group, the second channel consisted of a bipolar pair of electrodes recording the submental EMG signal. The subjects and patients were asked to close their eyes to reduce ocular and blink artefact. An Aspect A-1000 EEG monitor (software version 3.12; Aspect Medical Systems) was used to collect the EEG data. The processed BIS output was recorded at 5 s intervals on a laptop computer, while the raw EEG data were downloaded (sampling frequency 256 samples s1) onto a second computer for later analysis. Electrode impedance was less than 5000 ohms, and the low- and high-frequency filters were set at 0.5 and 70 Hz with the mains notch filter set at 50 Hz. This filtering applied to the processed output from the Aspect Monitor, but not the raw EEG output. All subsequent processing and analysis were done using purpose-written computer programs in Matlab (Matlab 5.3; Mathworks, Natick, Massachusetts, USA).
SE50d, BetaRatio and SynchFastSlow
The data were processed in 2 s epochs. The signal was low-pass filtered using a phase-preserving 12th-order elliptical digital filter (cutoff frequencies 4749 Hz, 100 dB roll-off, <2 dB ripple). Then, the signal was processed to reject gross artefacts by excluding epochs in which the absolute magnitude of the signal was greater than 200 µV. The BetaRatio and SynchFastSlow were obtained using the published formula described by Rampil.12 The exact algorithm for the calculation of the burst suppression ratio (QUAZI Suppression Index) is not in the public domain, and therefore the calculation was not done. In any case, the depth of anaesthesia was such that no burst suppression (as indicated by the Aspect Monitor) was achieved in any patient. The first time-derivative of the EEG signal was then taken and the spectral density of each 2 s epoch obtained (Hanning window). From this, the median frequency of the derivative of the signal (SE50d) could be calculated as the frequency which bisected the area under the derivative-power spectral density curve. The SE50d30Hz was also calculated using a version of the EEG signal that had been additionally digitally low-pass filtered to 30 Hz. This was done because it is common for commercial EEG systems to use this more severe filtering of higher frequencies. The 2 s epochs of all three variables were then smoothed using a moving five-point median window.
EMG influence
There has been some criticism that the high-frequency component (3047 Hz) of the BetaRatio is particularly susceptible to the influence of EMG interference. It is usual to consider the power spectrum of the observed scalp EEG signal (EEGscalp) as the sum of (i) the power spectrum of the true EEG signal (EEGtrue) and (ii) the power spectrum of the frontalis EMG signal (EMGfrontalis) (diagrammatically displayed in Fig. 2).
|
|
Statistical techniques
Unless stated otherwise, data are presented as mean (SD). The changes in the various EEG parameters at each time-point were compared statistically using paired t-tests with Bonferronis correction for multiple comparisons, if the data satisfied normality criteria. In the patient group the data were often skewed, and therefore the Wilcoxon test was used. Receiver operating curve (ROC) analysis was used to compare the ability of the EEG parameters to distinguish the awake and anaesthetized states and to determine the appropriate cutoff values.14 The anaesthetized state (at the time of surgery) was coded as positive. All the spectral powers were expressed as the 10·base 10 logarithm [i.e. dB relative to 1 (µV)2 Hz1]. It is important to remember that subtraction of logarithms is equivalent to division, and that a difference of one logarithmic unit is the equivalent of a tenfold difference in absolute value.
![]() |
Results |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
|
|
A statistical comparison of the changes in the variables is shown in Table 3. All the variables decreased significantly during induction of anaesthesia (P<0.001). In all the raw variables (i.e. other than the BIS), the most marked change occurred between the awake state at the start and the point of loss of consciousness. Using the area (SE) under the ROC curves, the SE50d (0.95 (0.12)), BIS (0.95 (0.12)) and BetaRatio (0.96 (0.12)) showed consistent and equivalent decreases when the values at the time of surgery were compared with those at the start. The SynchFastSlow was slightly less accurate (ROC area = 0.91 (0.12)). In contrast, the SE50d30Hz (0.80 (0.11)) and the 95% spectral edge frequency (0.52 (0.13)) were both significantly worse than the SE50d when comparing the area under ROC curve (P<0.005, t-test).
|
|
When estimating the proportion of the EEG signal that was contaminated by the frontalis EMG signal, we used the 3047 Hz frequency band, because that is the high-frequency component of the BetaRatio. Overall, the median (25th to 75th centile) of the ratio of EEGscalp(3047Hz): EMGfrontalis(3047hz) was 12.6 (2.121) dB. This means that, on average, the EEG signal in the 3047 Hz (i.e. low gamma) frequency band was more than eighteen (=101.26) times the magnitude of the EMG signal. However, the difference was more marked in the AWAKE state, for which the median ratio was 21.1 (14.727.5) dB. This corresponds to the EEG being over 100 times more powerful than the EMG in the gamma band. However, it must be noted that in the lowest decile of the patients the ratio was negative (i.e. the EMG power was greater than the EEG power). This indicates that, in some patients at least, there are inaccuracies in the assumptions used to derive the frontalis EMG amplitude.
![]() |
Discussion |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Compared with the BIS in our study, these alternative EEG indices are at an additional disadvantage because there was minimal artefact rejection in their preprocessing compared with the sophisticated artefact management used in the derivation of the BIS.12 All predictive results need further prospective testing on an independent group before they can be recommended for routine clinical use. It must also be emphasized that the EEG parameters were evaluated during the induction of general anaesthesia. The changes that may occur during reawakening at the end of anaesthesia may differ.
It seems plausible that the improved performance of the SE50d (over the spectral edge frequency and the SE50d30Hz) is due to: (i) the use of the first derivative, which has the effect of filtering out the fluctuating low-frequency artefact noise which, episodically, can dominate the true low-amplitude, high-frequency EEG power spectrum in the awake patient; and (ii) the fact that the SE50d emphasizes the importance of the desynchronized gamma rhythms, which are associated with the EEG of a subject in the attentive conscious state. The BetaRatio and the SE50d are similar in that they quantify the dominant EEG frequencies within a constrained frequency band. This includes the lower part of the gamma band and excludes the variable, dominant effects in the low frequencies (<10 Hz).
Schnider and colleagues, using the EEG as a monitor of propofol effect, have developed a variable called the semilinear canonical coefficient.15 In simple terms this is a development of the statistical method of canonical regression, which, by an iterative method, automatically derives the best combination of weights for each frequency bin to describe the changes in the EEG signal during the transition from the alert state to anaesthesia. The non-linear link function is required to allow for the variable biphasic EEG response that occurs in the stage of light anaesthesia. In keeping with traditional practice, Schnider and colleagues used low-pass filtering of the EEG signal to exclude frequencies above 30 Hz. The diagrams of changes in the semilinear canonical coefficient during induction of anaesthesia in their paper (Fig. 3) are very similar to that of the SE50d30 Hz in our paper (Fig. 4), showing a biphasic response. We suggest that this biphasic response may be attenuated or eliminated by the inclusion of higher frequencies in the analysed EEG signal.
There has been a number of studies using the auditory evoked steady-state response as a measure of depth of anaesthesia.16 17 These studies have shown a good correlation between increasing anaesthetic concentrations and attenuation of the auditory evoked steady-state response.18 Although the exact neurophysiological correlation between evoked (25 ms) 40 Hz activity and spontaneous EEG gamma-band activity is not entirely clear, there appears to be significant convergence: both are found to be relatively accurate indicators of the alert state.19 It is of interest that, in a study by Schraag,20 the area under the ROC curve (distinguishing the unconscious and conscious states) for the BIS (0.92) and their auditory evoked potential index (0.97) were very similar to the area under the ROC curves in our study. A simple estimator of the auditory response (the auditory evoked potential index) quantifies the response by summing the absolute magnitude of the first time-derivative of the signal, a procedure similar in some respects to the derivation of the SE50d.
The influence of changes in the frontalis EMG on the observed scalp EEG is problematical. Using our results, it would seem that, statistically, in about 90% of patients the EMG signal is a minor component of the observed scalp EEG signal. There is the important proviso that these conclusions are based on information obtained indirectly from the EMGsubmental and not directly from the (unmeasurable) EMGfrontalis. We have made the unproven assumption that, at least in the higher frequency ranges (>20 Hz), the slope of the power spectrum of the EMGfrontalis was similar to that observed in EEGsubmental. However, this problem is probably not clinically harmful, and is biased towards preventing intraoperative awareness in both possible scenarios. In most cases the effect of a high EMG spectral power is to flatten the BetaRatio, i.e. to make the patient appear more awake than he really is, and thus the anaesthetist would be tending to err towards increasing the amount of anaesthetic agent. Conversely, the EMGfrontalis may have a very steep spectral slope (corresponding to an EMG BetaRatio of perhaps 3.00) caused by noise in the form of an excessive low-frequency EMG component of the scalp EEG signal, which would suggest that the patient is asleep when in fact he is awake. In this case, if the patient had that amount of muscle activity he would be able to forcefully indicate his conscious state to the surgeon! It is fortunate that all EEG indices tend to become more reliable in the presence of muscle relaxation (as long as this effect is taken into account when setting the thresholds of EEG parameters used for determining awareness).
One of the main criticisms of the BIS has been that it is derived empirically and not based on theory and neurophysiological facts. This study may go some way towards clarifying why the BIS is effective. As has been previously stated,12 the BetaRatio is the important component of the BIS at light levels of anaesthesia. The success of this subcomponent in tracking the patients level of consciousness during induction of anaesthesia seems to indicate that, if the goal is to decide if the patient is aware or not, then the dominant frequency of the desynchronized oscillation is the important EEG feature. This phenomenon of blocking of the gamma rhythms is a feature of the commonly used GABAergic general anaesthetic agents, and is much less marked with excitatory anaesthetic agents such as nitrous oxide or ketamine, which tend to maintain the gamma-band EEG activity.21 22 This would explain why the BIS is not very sensitive when used during sedation with nitrous oxide.23 24 The utility of the SE50d and BetaRatio with nitrous oxide is unknown.
![]() |
Appendix 1 |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
BetaRatio = log10 [spectral power(3047 Hz)/
spectral power(1120 Hz)].
The SynchFastSlow is the logarithm of the ratio of the bispectral power in the waveband 4047 Hz to that in the band 0.547 Hz:
SynchFastSlow = log10[bispectral power(4047Hz)/
bispectral power(0.547Hz)].
It may be surmised that the success of the BIS may be due, in large part, to the emphasis on these higher frequencies extending into the gamma band. Although the bispectral index (BIS) is widely used as a monitor of the depth of anaesthesia, it has some disadvantages: (i) it is an empirically derived black box, with a complex non-linear algorithm that is difficult to relate to electrophysiological changes; (ii) the exact point at which individual patients lose consciousness occurs over a wide range of BIS values; (iii) it has a significant time lag, thus precluding individual titration to effect; and (iv) it does not reliably warn of impending arousal.
Almost all the published work on the higher-order spectra of the EEG signal in anaesthesia describes the changes in only the processed output of the Aspect Monitor: the BIS. Changes in the actual raw bicoherence and bispectrum of the EEG signal during natural sleep and epileptic seizures have been analysed by Bullock and colleagues.25 Because the basis of the algorithm for calculating the BIS is not freely available, it is difficult to interpret the observed changes in terms of known neurophysiological processes. To understand the BIS fully, it is necessary to dissect it into its various subcomponents as far as possible. The theoretical basis of the bispectrum of the EEG signal is difficult to visualize intuitively. It has been claimed that a significant bispectral power is indicative of non-linear (quadratic) interfrequency phase coupling.26 Strictly, this interpretation implies interactions among regular stationary cortical oscillators. It may be more realistic to consider the EEG as a stochastic (random) signal. In this case, the bispectral power may be better interpreted statistically as an estimate of the degree to which the signal is non-Gaussian, specifically its skewness.27 28 Measurement noise and the EMG signal are both approximately Gaussian in distribution, and thus should not appear in the bispectrum. Therefore, it is conceivable that the bispectral techniques are effective because they are effectively reducing the influence of unwanted extra-EEG noise rather than revealing information about the degree of neurological integration. Unfortunately, the interpretation of the bispectrum is further complicated by the fact that even a truly Gaussian time series will have a non-zero bicoherence, because the bispectrum is a biased estimator, being distributed approximately as a 2 distribution.29 Thus, it could be argued that the inclusion of complex bispectral measures is not important as long as the changes in high-frequency content are included in whatever EEG variable is chosen. If this is true, we could expect that similar information could be obtained without the necessity of using higher-order spectral techniques, by the use of a simpler, cut-down component of the BISthe BetaRatio.
![]() |
Appendix 2 |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
The ratio of submental to frontalis EMG power in the 3047 Hz band equals the ratio in the 5585 Hz band (using the symbols shown in Fig. 3):
A/C = D/E(1)
where A is the 3047 Hz submental power (EMGsubmental(3047Hz)), C is the 3047 Hz frontalis power (EMGfrontalis(3047Hz)) (unknown), D is the 5585 Hz submental power (EMGsubmental(5585Hz)) and E is the 5585 Hz frontalis power (EMGfrontalis(5585Hz)). C is unknown but can be calculated from equation (1).
Actually we want to estimate the ratio B/C, where B is the 3047 Hz total EEG power (EEGscalp(3047Hz)).
From equation (1),
B/C = (B·D)/(A·E)(3)
Taking logs, and expressing in decibels,
10log10 (B/C) = 10log10 (B) + 10log10 (D)
10log10 (A) 10log10 (E).
This will allow us to quantify the extent to which the total observed EEG power in the 3047 Hz band (EEGscalp(3047Hz)) has been contaminated by frontalis EMG power. All the terms on the right-hand side of the equation are measurable experimentally.
![]() |
References |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
2 Billard V, Gambus PL, Chamoun N, Stanski SL, Shafer SL. A comparison of spectral edge, delta power, and bispectral index as EEG measures of alfentanil, propofol, and midazolam effect. Clin Pharmacol Ther 1997; 61: 4558[ISI][Medline]
3 Desmet JE, Tomberg C. Transient phase-locking of 40-Hz electrical oscillations in prefrontal and parietal human cortex reflects the process of conscious somatic perception. Neurosci Lett 1994; 168: 1269[ISI][Medline]
4 Menon V, Freeman WJ, Cutillo BA, Desmond JE. Spatio-temporal correlations in human gamma band electrocorticograms. Electroencephalogr Clin Neurophysiol 1996; 98: 89102.[ISI][Medline]
5 Rodriguez E, George N, Lachaux J-P, Martinerie J, Renault B, Varela F. Perceptions shadow: long distance synchronization of human brain activity. Nature 1999; 397: 4304[ISI][Medline]
6 Franken P, Dijk DJ, Tobler I, Borbely AA. High-frequency components of the rat electrocorticogram are modulated by the vigilance states. Neurosci Lett 1994; 167: 8992.[ISI][Medline]
7 Gross DW, Gotman J. Correlation of high-frequency oscillations with the sleepwake cycle and cognitive activity in humans. Neuroscience 1999; 94: 100518[ISI][Medline]
8 Tiitinen H, Sinkkonen J, Reinikaien K, Alho K, Lavikalnen J. Selective attention enhances the auditory 40-Hz transient response in humans. Nature 1993; 364: 5960[ISI][Medline]
9 Vaadia E, Haalman I, Abeles M, et al. Dynamics of neuronal interactions in monkey cortex in relation to behavioural events. Nature 1995; 373: 51521[ISI][Medline]
10 Whittington MA, Traub RD, Jefferys JGR. Synchronized oscillations in interneuron networks driven by metabotropic glutamate receptor activation. Nature 1995; 373: 6125[ISI][Medline]
11 Pomfrett CJD. Heart rate variability, BIS and depth of anaesthesia. Br J Anaesth 1999; 82: 65962
12 Rampil IJ. A primer for EEG signal processing in anaesthesia. Anesthesiology 1998; 89: 9801002[ISI][Medline]
13 Williams ML, Sleigh JW. Auditory recall and response to command during recovery from propofol anaesthesia. Anaesth Intens Care 1999; 27: 2658[ISI][Medline]
14 Hanley JA, McNeil BJ. The meaning and use of the area under the receiver operating characteristic (ROC) curve. Radiology 1982; 143: 2936[Abstract]
15 Schnider TW, Minto CF, Shafer SL, Gambus PL, Andresen C, Goodale DDS, Youngs EJ. The influence of age on propofol pharmacodynamics. Anesthesiology 1999; 90: 150216[ISI][Medline]
16 Dutton RC, Rampil IJ, Chortkoff BS, Eger EI 2nd. Forty-hertz midlatency auditory evoked potential activity predicts wakeful response during desflurane and propofol anesthesia in volunteers. Anesthesiology 1999; 91: 120920[ISI][Medline]
17 Gilron I, Marcantoni W, Varin F. 40 Hz auditory steady-state response and EEG spectral edge frequency during sufentanil anaesthesia. Can J Anaesth 1998; 45: 11521[Abstract]
18 Plourde G, Villemure C, Fiset P, Bonhomme V, Blackman SB. Effect of isoflurane on the auditory steady-state response and on consciousness in human volunteers. Anesthesiology 1998; 89: 84451[ISI][Medline]
19 Plourde G. Evoked potentials and 40-Hz oscillations. Anesthesiology 1999; 91: 11879[ISI][Medline]
20 Schraag S, Gajraj R, Kenny GNC, Georgieff M. The performance of electroencephalogram bispectral index and auditory evoked potential index to predict loss of consciousness during propofol infusion. Anesth Analg 1999; 89: 13115
21 Clark DL, Rosner BS. Neurophysiological effects of general anaesthetics. I. The electroencephalogram and sensory evoked responses in man. Anesthesiology 1973; 38: 56482[ISI][Medline]
22 Plourde GP, Baribeau J, Bonhomme V. Ketamine increases the amplitude of the 40-Hz auditory steady-state response in humans. Br J Anaesth 1997; 78: 5249
23 Barr G, Jakobsson JG, Owall A, Anderson RE. Nitrous oxide does not alter bispectral index: study with nitrous oxide as sole agent and as adjunct to i.v. anaesthesia. Br J Anaesth 1999; 82: 82730
24 Rampil IJ, Kim JS, Lenhardt R, Negishi C, Sessler DI. Bispectral EEG index during nitrous oxide administration. Anesthesiology 1998; 89: 6717[ISI][Medline]
25 Bullock TH, Achimowicz JZ, Duckrow RB, Spencer SS, Iragui-Madoz VJ. Bicoherence of intracranial EEG in sleep wakefulness and seizures. Electroencephalogr Clin Neurophysiol 1997; 103: 66178[ISI][Medline]
26 Sigl JC, Chamoun NC. An introduction to bispectral analysis for the electroencephalogram. J Clin Monit 1994; 10: 392404[ISI][Medline]
27 Mendel JM. Tutorial on higher order statistics (spectra) in signal processing and system theory: theoretical results and some applications. Proc IEEE 1991; 79: 278305[ISI]
28 Mclaughlin S, Stogioglou A, Fackrell J. Introducing higher order statistics (HOS) for the detection of nonlinearities. http://www.amsta.leeds.ac.uk/Aplii..dir/issue2/hos_intro.html 1995
29 Elgar S. Statistics of bicoherence and biphase. J Geophys Res 1989: 94; 109938[ISI]