Secondary Logo

Journal Logo

A Procedural Electroencephalogram Simulator for Evaluation of Anesthesia Monitors

Petersen, Christian Leth PhD*; Görges, Matthias PhD*†; Massey, Roslyn BSc (cand.)*; Dumont, Guy Albert PhD; Ansermino, J. Mark MBBCh, MSc (Inf), FFA (SA), FRCPC*

doi: 10.1213/ANE.0000000000001506
Technology, Computing, and Simulation: Original Clinical Research Report

BACKGROUND: Recent research and advances in the automation of anesthesia are driving the need to better understand electroencephalogram (EEG)–based anesthesia end points and to test the performance of anesthesia monitors. This effort is currently limited by the need to collect raw EEG data directly from patients.

METHODS: A procedural method to synthesize EEG signals was implemented in a mobile software application. The application is capable of sending the simulated signal to an anesthesia depth of hypnosis monitor. Systematic sweeps of the simulator generate functional monitor response profiles reminiscent of how network analyzers are used to test electronic components.

RESULTS: Three commercial anesthesia monitors (Entropy, NeuroSENSE, and BIS) were compared with this new technology, and significant response and feature variations between the monitor models were observed; this includes reproducible, nonmonotonic apparent multistate behavior and significant hysteresis at light levels of anesthesia.

CONCLUSIONS: Anesthesia monitor response to a procedural simulator can reveal significant differences in internal signal processing algorithms. The ability to synthesize EEG signals at different anesthetic depths potentially provides a new method for systematically testing EEG-based monitors and automated anesthesia systems with all sensor hardware fully operational before human trials.

Published ahead of print July 26, 2016.

From the *Department of Anesthesiology, Pharmacology & Therapeutics, University of British Columbia, Vancouver, British Columbia, Canada; Pediatric Anesthesia Research Team, Child and Family Research Institute, Vancouver, British Columbia, Canada; and Department of Electrical and Computer Engineering, University of British Columbia, Vancouver, British Columbia, Canada.

Published ahead of print July 26, 2016.

Accepted for publication January 19, 2016.

Funding: Supported in part by the Society for Technology in Anesthesia Fresenius Research Award.

Conflict of Interest: See Disclosures at the end of the article.

Reprints will not be available from the authors.

Address correspondence to Christian Leth Petersen, PhD, Pediatric Anesthesia Research Team, Room V3-350, 950 West 28th Ave, Vancouver, BC, Canada V5Z 4H4. Address e-mail to

Anesthesia delivery guided by a measured physiological end point, such as depth of hypnosis, offers the ability to overcome interpatient variability. Currently, depth of hypnosis measurements are hampered by the varied signal extraction and processing techniques employed by monitor manufacturers, as well as the limited understanding of various anesthetic drug effects on electroencephalogram (EEG), the most commonly used physiological signal for determining the depth of hypnosis.

A number of different EEG-based monitors are commercially available to estimate the depth of hypnosis. These monitors use different processing algorithms, many of them undisclosed, to provide a single number that represents the depth of hypnosis. These algorithms are typically developed and validated using different clinical scores in different clinical contexts.

EEG-based anesthesia monitor variability is difficult to test in vivo because only 1 set of monitor electrodes can practically be placed on the forehead of a patient while adhering to manufacturer recommendations. This problem can be addressed, to some extent, by replaying the prerecorded EEG to multiple monitors1 or by feeding signals into multiple monitors at once using nonstandard electrode placements.2

Testing of EEG-based feedback systems with full electrical sensor emulation and all hardware sensor interfaces online (hardware-in-the-loop) requires an additional way to generate EEG corresponding to a given depth of hypnosis level. Replaying stitched/looped prerecorded EEG segments at the desired depth of hypnosis level causes spurious readings on the monitors caused by the discontinuities between segments.

In this report, a procedural EEG generator that uses a simple 1-parameter model to simulate EEG during anesthesia is described and used as a functional analyzer to explore differences between anesthesia monitors under steady-state (stable) conditions. Three common monitors were tested against the simulator: Entropy (GE Healthcare, Little Chalfont, UK), NeuroSENSE (NeuroWave Systems, Cleveland, OH), and BIS (Medtronic, Galway, Ireland). These monitors output a depth of hypnosis index on a scale of 0 to 100. The scales are not standardized, although the manufacturers of all 3 monitors recommend an appropriate depth of general anesthesia as 40 to 60, with lower numbers indicating increasing depth of hypnosis. The algorithms used in the 3 monitors are very different and not fully disclosed:

  • (1) The Entropy monitor uses the response entropy index.3 This is based on time-balanced spectral entropy calculations of signals from 0.8 to 47 Hz, the electromyography (EMG)-dominated portion of the range.4 This range, and the EEG-dominated state entropy range of 0.8 to 32 Hz, are used in conjunction to determine the depth of hypnosis.5
  • (2) The NeuroSENSE monitor uses the WAVcns index using the probability density functions (PDFs) of the wavelet transform coefficients of an awake EEG waveform and an isoelectric wave as references.6 The monitor calculates the PDF corresponding to the incoming EEG and compares it with the reference PDFs to generate the depth of hypnosis.7
  • (3) The BIS monitor uses a complex algorithm that feeds multiple time, frequency, and higher-order statistical parameters extracted from the EEG signal through a weighted range-dependent decision algorithm to arrive at a depth of hypnosis.8
Back to Top | Article Outline


The development of an empirical model of temporal EEG structure, without making assumptions about the workings of any individual monitor, is the central concept. Considering that the monitors are all attempting to map the same anesthetic effect to similar nominal scales, the monitors would be expected to generate similar output when presented with the same signal based on this model.

Consumer-grade audio digital-to-analog converters can be readily used to simulate the EEG signal. In fact, the standard audio port on smartphones opens an opportunity of implementing the EEG simulator as a cross-platform LambdaNative9 ( smartphone app for iOS and Android.10 This reduces the hardware requirements and makes it possible to disseminate the technology easily through global app stores.

It is important for the simulator to mimic the electrical properties of the scalp;11 otherwise, the attached anesthesia monitor will report electrode errors and fail to accept the signal from the simulator. This is accomplished by sending the audio signal through an electrical impedance network (Figure 1A) that also acts as a voltage divider (approximately 1:10,000). The full electrical schematic for this interface is available online.12

Figure 1.

Figure 1.

Under ideal conditions, the goal would be to generate an EEG signal based on a given depth of hypnosis with a simple 1-parameter model that uses only a single independent variable. It is well known that the EEG changes from a low-amplitude, high-frequency signal at wakefulness to a high-amplitude, low-frequency signal during anesthesia (Figure 1B). This behavior is the macroscopic reflection of complex changes in the neurological state, which is difficult to model on the microscopic neuronal network level.13 We opted instead for an empirical approach aiming to replicate the observed temporal behavior by using an analogy to terrain morphology. Physical terrain is a macroscopic result of complex microscopic critical processes, the procedural rendering of which was developed decades ago.14

Gradient noise texture generators, now ubiquitous in modern computer graphics, are based on a normalized pseudorandom smoothing function derived from interpolation of a seeded random number series. Such noise generators are commonly used to represent naturally occurring 1/f noise, which is similar in structure to the EEG. We can apply this gradient noise directly to generate an EEG signal at different levels of hypnosis simply by scaling the noise function appropriately in time and amplitude:

where α is a constant scaling factor, t is time, GradientNoise is the gradient noise function, here in the form of 1-dimensional Perlin noise,14 and simulation parameter k is a factor related to the depth of hypnosis. The value of α is not critical because the EEG amplitude does not impact anesthesia monitor readings directly. α is determined by adjusting a sinusoidal output waveform to 100 μV peak-to-peak amplitude on the anesthesia monitor display in a calibration mode of the software application. The mapping between k and the target commercial anesthesia monitor depth of hypnosis index reveal information about the internal algorithms of the monitors. Provided that this mapping is single valued (monotonic), its reverse acts as a calibration to each monitor that can serve to drive closed-loop simulations.

Figure 2.

Figure 2.

Approval from the University of British Columbia and Children’s and Women’s Health Centre of British Columbia research ethics board (H15-03244) for secondary use of previously collected raw bilateral EEG data during IV anesthesia (propofol and remifentanil) was obtained to compare the morphology of the real EEG signal and the procedural simulator output. The signals were evaluated using cross-recurrence plots15,16 (Figure 2) because these plots help visualize the dynamic evolution of the bipolar signals and reveal simultaneous occurrences of similar states in the 2 channels. Figure 2 illustrates how the simple procedural model replicates the primary temporal dynamics of the EEG signal in the case of wakefulness (depth of hypnosis = 90) (Figure 2A and 2B) and anesthesia (depth of hypnosis = 50) (Figure 2C and 2D).

Back to Top | Article Outline


The monitor readings were compared at steady state to ensure consistent and reproducible results, with a time delay of 30 seconds (Figure 3A) and 60 seconds (Figure 3B) after each change in k. The results are reproducible and were verified against 2 different monitor units of different manufacturing year and version. Only the BIS monitors showed noticeable difference in behavior between the 2 delays.

Figure 3.

Figure 3.

The Entropy and NeuroSENSE signals are coincidental at a high depth of hypnosis indices, and both monitors appear to be relatively insensitive to changes in the simulator signal in this regime. In the clinically relevant range of 30 to 70, the Entropy and the NeuroSENSE monitor outputs are approximately linearly related to the model parameter k. The slopes of the linear portions are similar, yet the curves are laterally shifted in relation to each other. At a low depth of hypnosis indices, the NeuroSENSE sensitivity appears to taper off, whereas the Entropy monitor enters another linear portion with a different slope.

The BIS response is nonmonotonic and appears different from both the Entropy and the NeuroSENSE. The BIS depth of hypnosis readings never reach >85, and in the upper range, the BIS monitor shows a clear reproducible 2-state oscillation in which the monitor toggles between levels that are approximately 15 points apart. Significant hysteresis is present in this region. The increasing parameter sweep of the noise generator gives BIS readings primarily in the upper of the 2 states, whereas the decreasing parameter sweep gives results primarily in the lower state. The clinically relevant range is nonlinear and dominated by a plateau at a depth of hypnosis = 50 with reproducible oscillations. At depth of hypnosis values <50, the BIS readings are very similar to those of Entropy.

Back to Top | Article Outline


We consider the depth of hypnosis performance with respect to 3 intervals of clinical significance: light, normal, and deep anesthesia:

  • (1) Depth of hypnosis >70 (light anesthesia): We attribute the low maximum readings for the BIS to the lack of wakeful EMG in the simulator model. The nonlinear behavior >70 could also be influenced by the lack of EMG. This region has abrupt change points, which are likely caused by transitions between the multiple internal algorithms of the monitor. This finding suggests that there might be circumstances in which the BIS readings can be unstable during induction and/or recovery.
  • (2) 70 > depth of hypnosis > 30 (normal anesthesia): The linearity observed in the clinically relevant section of NeuroSENSE and Entropy readings was expected. However, there is a significant unexpected offset between them, suggesting that for a given level of anesthesia, Entropy would tend to read higher than NeuroSENSE. For the BIS, the large reproducible variability, especially near k = 4 in Figure 2A, and broad plateau in depth of hypnosis values around 50 are concerning because this is the central region of clinical interest where a strong and uniform response is desirable.
  • (3) Depth of hypnosis <30 (deep anesthesia): The observations at a low depth of hypnosis indices are likely shadowed by the presence of burst suppression (BS) in a real clinical situation. The current simulator model does not include BS, which would be detected by the monitors and affect the low readings. Although BS is easily added to the simulator, it would trigger separate states in the monitors and confuse our purpose of testing the fundamental depth of hypnosis response in each monitor.

Overall, the NeuroSense had the least noise scatter, with the Entropy consistently having more noise across the range. The BIS more closely approximated the parameter shift in the noise generator than the Entropy in some regions, but it exhibited significant noise at low depth of hypnosis values.

The substantial differences observed between the 3 evaluated monitors highlight the importance of depth of hypnosis signal processing algorithms. In particular, some of the variance observed in the BIS monitor may be attributable to the nonlinear aspects of the BIS algorithm (the bispectrum). To fully assess the differences, it would be necessary for manufacturers to disclose the exact implementations used in their monitors. Alternatively, a completely open algorithm can be developed independently from systematic analysis (eg, the above). Such a standardized depth of hypnosis measure could support a new generation of robust automated anesthesia delivery systems.

The audio interface schematic of the EEG simulator is freely available under a Community Commons–type license, and we plan to make the smartphone app freely available to the anesthesia community as well.12 The simulator can serve the community as a new tool to improve safety in anesthesia and to further the development of robust agent delivery systems for the future of anesthesia.

We have compared the output from anesthesia monitors in a systematic way using a procedural model. No direct clinical conclusions can be drawn from these findings until they are verified against clinical data because the model lacks EMG, and BS and is not derived from actual EEG. This will be addressed in a future study.

Back to Top | Article Outline


Name: Christian Leth Petersen, PhD.

Contribution: This author helped design the study, analyze the data, and write the manuscript.

Conflicts of Interest: Christian Leth declares no conflicts of interest.

Name: Matthias Görges, PhD.

Contribution: This author helped design the study, conduct the study, analyze the data, and write the manuscript.

Conflicts of Interest: Matthias Görges declares no conflicts of interest.

Name: Roslyn Massey.

Contribution: This author helped conduct the study and write the manuscript.

Conflicts of Interest: Roslyn Massey declares no conflicts of interest.

Name: Guy Albert Dumont, PhD.

Contribution: This author helped analyze the data and write the manuscript.

Conflicts of Interest: Guy Albert Dumont has equity interest in BioNova Technologies.

Name: J. Mark Ansermino, MBBCh, MSc (Inf), FFA (SA), FRCPC.

Contribution: This author helped analyze the data and write the manuscript.

Conflicts of Interest: J. Mark Ansermino declares no conflicts of interest.

This manuscript was handled by: Maxime Cannesson, MD, PhD.

Back to Top | Article Outline


1. Pilge S, Kreuzer M, Karatchiviev V, Kochs EF, Malcharek M, Schneider G. Differences between state entropy and bispectral index during analysis of identical electroencephalogram signals: a comparison with two randomised anaesthetic techniques. Eur J Anaesthesiol. 2015;32:354365.
2. Bresson J, Gayat E, Agrawal G, et al. A randomized controlled trial comparison of NeuroSENSE and bispectral brain monitors during propofol-based versus sevoflurane-based general anesthesia. Anesth Analg. 2015;121:11941201.
3. Davidson AJ, Huang GH, Rebmann CS, Ellery C. Performance of entropy and Bispectral Index as measures of anaesthesia effect in children of different ages. Br J Anaesth. 2005;95:674679.
4. Viertiö-Oja H, Maja V, Särkelä M, et al. Description of the Entropy algorithm as applied in the Datex-Ohmeda S/5 Entropy module. Acta Anaesthesiol Scand. 2004;48:154161.
5. Pilge S, Kreuzer M, Karatchiviev V, Kochs EF, Malcharek M, Schneider G. Differences between state entropy and bispectral index during analysis of identical electroencephalogram signals: a comparison with two randomised anaesthetic techniques. Eur J Anaesthesiol. 2015;32:354365.
6. Bibian S, Dumont GA, Zikov T. Dynamic behavior of BIS, M-entropy and neuroSENSE brain function monitors. J Clin Monit Comput. 2011;25:8187.
7. Zikov T, Bibian S, Dumont GA, Huzmezan M, Ries CR. Quantifying cortical activity during general anesthesia using wavelet analysis. IEEE Trans Biomed Eng. 2005;11:535557.
8. Rampil IJ. A primer for EEG signal processing in anesthesia. Anesthesiology. 1998;89:9801002.
9. Petersen CL, Görges M, Dunsmuir D, Ansermino JM, Dumont GA. Experience report: functional programming of mHealth applications. Proceedings of the 18th ACM SIGPLAN International Conference on Functional programming. 2013:357362.
10. Petersen CL, Ansermino JM, Dumont GA. Towards a depth of anesthesia simulator. Anesth Analg. 2016;122(5S suppl):3536.
11. Kreuzer M, Kochs EF, Pilge S, Stockmanns G, Schneider G. Construction of the electroencephalogram player: a device to present electroencephalogram data to electroencephalogram-based anesthesia monitors. Anesth Analg. 2007;104:135139.
13. Olofsen E, Dahan A. Big brain, small world? Anesthesiology. 2015;122:811.
14. Perlin K. An image synthesizer. SIGGRAPH ‘85 Proceedings of the 12th Annual Conference on Computer Graphics and Interactive Techniques. 1985:287296.
15. Eckmann JP, Oliffson Kamphorst S, Ruelle D. Recurrence plots of dynamical systems. Europhys Lett. 1987;4:973977.
16. Zbilut JP, Webber CL. Detecting deterministic signals in exceptionally noisy environments using cross-recurrence quantification. Phys Lett A. 1998;246:122128.
Copyright © 2016 International Anesthesia Research Society