Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Through the recording of scalp surface electrical activity, electroencephalography (EEG) is unique as a noninvasive tool to explore the functioning brain. The obtained complex tracings correspond to the summation of the activity of several underlying neural assemblies, and they contain a huge amount of information. The time resolution of this tool is excellent, allowing an almost immediate assessment of brain function, while the space resolution is acceptable when multiple sites of recording are used. The main difficulty resides in extracting the contained information. Several modes of tracing analysis exist, from simple to highly complex, providing corresponding basic to system-integrated information. EEG can also be combined with other techniques, such as transcranial magnetic stimulation (TMS), or functional magnetic resonance imaging (fMRI), to further enhance the understanding of communication between brain regions and information integration by the brain. EEG repercussions of peripheral sensory stimulation can also be studied in the form of evoked potentials, which can be evidenced after repeated stimuli and averaging of EEG waves. Their amplitude and latency can provide very useful information on peripheral afferences processing by the central nervous system. Several of those EEG modalities have given birth to commercially available monitors that are of current use in clinical practice. We will here review all those aspects of EEG, except for the evoked potentials that are extensively described in Chapter VI.
EEG consists in recording electric potential differences between electrodes placed on the scalp surface. Those potentials differences are usually said to be generated by the activity of cortical pyramidal neurons, oriented perpendicularly to the cortical surface ( Fig. 4.1 ). The synchronous firing of large groups of neurons generates detectable potential differences. These activities are influenced by cortico-cortical or cortico-subcortical interactions, notably with the thalamus. The obtained final complex wave corresponds to the summation of all those activities.
After digitalization (see Depth of Anesthesia Monitoring section), EEG can be analyzed through several ways ( Table 4.1 ). The most common technique is the visual inspection of the tracing by experienced electrophysiologists, leading to a syntactic or semantic description of the wave shape, amplitude, and frequency. Several EEG patterns can be visually recognized that correspond to different clinical conditions, including sleep stages, focal or generalized seizure, focal or global brain suffering, encephalopathy, brain death, or anesthesia. Such a visual analysis necessitates strong experience, and it is not within reach of all clinicians. In addition to the easy storage of information, the digitalization of the EEG offers several advantages, including post hoc mathematical and statistical manipulation of the tracing to improve filtering, extracting nonvisible parameters that are potentially interpretable by the nonelectrophysiologist, identifying sources of activity, characterizing interactions between brain regions, and making mechanistic inferences regarding brain functions.
Type | Name | Domain | Possible clinical applications | |
---|---|---|---|---|
Visual inspection | EEG alone | Wave shape, frequency, amplitude | Pattern recognition | Sleep stages Focal or generalized seizure Focal or global brain suffering Encephalopathy Brain death DHOA |
Digitized EEG | Zero crossing frequency | Time | Rare | |
α, β; θ, δ; and γ power | Frequency | DHOA | ||
Peak frequency | Frequency | DHOA | ||
Median frequency | Frequency | DHOA | ||
Spectral edge frequency | Frequency | DHOA | ||
Discrete wavelet transform, eigenvector analysis | Time and frequency | DHOA | ||
Coherence | Frequency | DHOA | ||
Bispectrum or phase synchrony | Frequency | DHOA | ||
Shannon or symbolic entropy (amplitude) | Disorder | DHOA | ||
Spectral entropy (power spectrum) | Disorder | DHOA | ||
Approximate, sample, and permutation entropy | Stability in time | DHOA | ||
Autoregressive modeling | Stability/complexity of activity | Mechanistic | ||
Granger causality and symbolic transfer entropy | Functional interactions between brain regions, effective connectivity | Mechanistic | ||
Dynamic causal modeling, mean-field modeling | Effective connectivity, mechanistic | Mechanistic | ||
Phase-amplitude coupling of frequency bands | Connectivity | Mechanistic | ||
Lempel-Ziv complexity, amplitude coalition entropy, synchrony coalition entropy | Effective connectivity and activity complexity | Mechanistic | ||
EEG + TMS | Perturbational complexity index | Effective connectivity, information differentiation and integration | Mental content assessment | |
EEG + fMRI | Functional and effective connectivity | Mechanistic |
Time and frequency domain analyses of the complex EEG wave are considered first-level analyses of the processed EEG. Time domain analysis describes the tracing in time, with parameters such as the zero-crossing frequency, which is the number of times the EEG wave crosses the baseline per unit of time. Frequency domain analyses first require the dissection of the complex EEG wave into all its individual sinusoidal components. This can be done using a Fourier transformation. The result is given in terms of several frequency bands and their relative contribution to the tracing total power, namely the power spectrum of the EEG. Classical frequency bands encompass the alpha (α; from 8 to 12 Hz), beta (β; 12–30 Hz), delta (δ; 0.5–4 Hz), and theta (θ; 4–8 Hz) bands. The gamma (γ; 25–100 Hz) band corresponds to higher frequencies. Several parameters can be derived from the power spectrum. For example, the peak frequency is the one with the highest power, the median frequency corresponds to the frequency where 50% of the power is achieved by lower frequencies, and 50% by higher ones, and the spectral edge frequency corresponds to the frequency where a specified percentage of the power (generally 90% or 95%) is achieved by lower frequencies.
Discrete wavelet transform and eigenvector analysis are other modes of EEG tracing dissection that allow describing the tracing in the time and frequency domain. The method consists in extracting coefficients that represent the correlation between a standard or “mother” wavelet and a localized section of the signal. It is a generalization of Fourier transformation that adds time elements to the frequency analysis, using signal windows of variable size. Wavelet analysis facilitates the analysis of rapidly changing signals.
Time and frequency domain analyses of the EEG of higher complexities have been proposed. For example, coherence measures synchronization between brain regions at a given frequency range, such as the gamma band. Other examples are the bispectrum or phase synchrony, corresponding to phase correlation between different frequency components.
The EEG is a complex signal, and one approach to its analysis can be the estimation of the disorder within the signal, or, in other words, its entropy. This disorder will vary as a function of brain state, being lower when brain activity is lower and vice-versa. EEG entropy can concern the amplitude (Shannon or symbolic entropy) or the power spectrum (spectral entropy). EEG entropy can also correspond to the stability of the EEG in time. When highly stable, such as during anesthesia, future amplitudes can be predicted with accuracy from previous amplitudes, and entropy is low. Approximate, sample, permutation, and multiscale entropies are examples of such parameters estimating EEG stability. Another way of estimating EEG stability/complexity uses autoregressive modeling methods.
The fact that anesthetic agents have effects on the EEG has been known for a long time, but changes in the EEG pattern as a function of anesthesia depth are not necessarily easy to recognize visually. In the 1990s, EEG-derived indexes were proposed to simplify EEG interpretation. Those indexes were initially designed to assess the hypnotic component of anesthesia, or importance of consciousness alteration. The basic principle governing the calculation of such indexes is the extraction of processed EEG parameters, and their introduction into a mathematical algorithm whose output is a dimensionless number between 0 and 100 or a letter. Usually, 0 corresponds to the deepest stage, or isoelectric EEG, and 100 to the wake state. Several depth of the hypnotic component of anesthesia (DHOA) monitors are currently commercially available.
BIS™ (Medtronic, Covidien, Minneapolis, MN, USA) is the first ever commercialized DHOA monitor, and the most widely studied. Its algorithm takes account of parameters such as the activity in the β frequency range, synchronized fast-slow activity, bispectrum, and burst suppression (alternating bursts and flat signal).
As described in Disorder Within the Electroencephalogram , spectral entropy measures the disorder into the power spectrum of the EEG. The obtained value is then normalized on a 0 to 100 scale, and is thus independent of the frequency and amplitude scale of the EEG. The mathematical algorithm of the GE Datex-Ohmeda Entropy module generates two numbers, the Response Entropy (RE), and the State Entropy (SE). SE is calculated over a frequency range between 0.8 and 32 Hz, which grossly corresponds to EEG activity. RE covers frequencies between 0.8 and 47 Hz, which involves the electromyographic activity of facial muscles in addition to EEG, although with overlap. Normalized SE values range between 0 and 91 and are presumed to reflect the hypnotic component of anesthesia. RE is always higher or equal to SE, and it ranges between 0 and 100. The occurrence of noxious stimulation in the absence of adequate antinociception provokes a facial muscles' jerk, hence an increase in RE. Consequently, the gradient between RE and SE is proposed to reflect the intraoperative balance between nociception and antinociception (see also Basic Principles , Nociception-anti-nociception Balance Monitoring , The Gradient Between State and Response Entropy ).
The development of Narcotrend™ (Narcotrend-Gruppe, Hannover, Germany) started with the identification of 15 different EEG patterns recorded during corresponding deepening stages of anesthesia. Each of those patterns was then further characterized by a set of processed EEG parameters extracted from the power spectrum, evaluating entropy, or derived from autoregressive analyses. Again, the output is a number between 0 and 100, in addition to a letter specifying the EEG pattern. The letter A indicates a wake state, B0 to B2 sedation, C0 to C2 light anesthesia, D0 to D2 general anesthesia, E0 to E2 general anesthesia with deep hypnosis, and F0 to F1 general anesthesia with increasing burst suppression.
The PSI (SedLine™ monitor, Masimo Corp., Irvine, CA, USA) output is also a number between 0 and 100, derived from the relative activity in specific frequency bands, interhemispheric coherence, and anterior to posterior frequency and phase relationships. It necessitates 4 EEG channels, and its development was mainly based on the anesthesia-induced frontal changes in the EEG spectrum.
CSI (Cerebral State Monitor™, Danmeter, Odense, Denmark) calculation involves the activity in the α and β frequency range, the difference between those two activities, and the importance of burst suppression.
For the WAV CNS (NeuroSENSE™ monitor, NeuroWave Systems Inc., Cleveland Heights, OH, USA), time and frequency domain information contained in the EEG is transposed into wavelet coefficients (see Time and Frequency Domain Analyses ). It is highly stable during a steady-state anesthesia and has the shortest time needed for calculation among the commercially available DHOA monitors. Hence, it has a very short reaction time when conditions are changing during anesthesia.
The aepEX™ monitor (Medical Device Management Ltd., Essex, UK) is slightly different from the other available DHOA monitors regarding the parameters that enter its algorithm. Following earphones-delivered auditory 7 Hz clicks, the device extracts middle latency auditory evoked potentials using a moving averaging window, and it calculates a 0 to 100 index based on their latency and amplitude. Indeed, amplitude decreases and latency increases with depth of hypnosis.
The qCON (qCON2000™, Quantium Medical S.L., Mataro (Barcelona), Spain) uses an Adaptive Neuro Fuzzy Inference System applied to a combination of a set of EEG frequency bands. The monitor also calculates the qNOX index, based on the same technology but devoted to the monitoring of the nociception–antinociception balance (see Basic Principles , Nociception-anti-nociception Balance Monitoring , The qNOX ). The output number ranges between 0 and 99 .
The development of the cortical state and its more robust derivative, the composite cortical state (CCS), was based on a mathematical EEG model, which aims at describing the influence external input on cortical activity. CCS is proposed to assess the response of the cortex to an arbitrary stimulus, hence DHOA, in combination with another index, the cortical input (CI, see The Cortical Input .), which quantifies the magnitude of the input to the cortex, and hence should better reflect the effect of antinociceptive medications. Up to now, CCS has been relatively poorly evaluated in clinical practice, but it seems to nicely correlate with DHOA.
Whereas the EEG is the measurement of the spontaneous electrical activity of the cerebral cortex, the evoked potentials are the measurement of the EEG response to a stimulus. They can be measured without patient cooperation and are therefore useful for diagnosis in the operating room under general anesthesia or in the intensive care unit. Several evoked responses can be used for the assessment of the nervous system integrity, including somatosensory-evoked responses (SSEP) obtained following repeated stimulation of the somatosensory system, auditory evoked responses, and visual evoked responses. Other modalities do not look at the EEG response, but rather to the peripheral response after central stimulation. For example, motor-evoked potentials are recorded in muscles following the transcranial electrical or magnetic stimulation of the motor cortex that activates the descending motor pathways. All those evoked responses modalities are discussed in detail in Chapter 6 and will not be covered here, except for some of them that are useful for the assessment of DHOA and the nociception–antinociception balance.
During anesthesia, individually tailoring the balance between nociception, corresponding to the unconscious consequences of noxious stimulation on the body, and antinociception, corresponding to the administration of antinociceptive medications such as opioids, is important to avoid unwanted movements or excessive autonomic responses to surgical stimulation, and to avoid overdose and its potential side effects, such as prolonged ventilation, escape phenomenon, or postoperative hyperalgesia. Several devices have recently been proposed to better assess this nociception–antinociception balance, or NAN balance. Most of them are based on parameters that reflect changes in the autonomic nervous system in response to noxious stimulation. By nature, those parameters may be influenced by factors that have nothing to do with nociception, such as the intravascular volume status, temperature, or vasoactive medications. Some authors have looked at more central witnesses of nociception processing, to bypass the autonomic nervous system confounding factors. Although nociception is mainly mediated by subcortical circuits located in the spinal cord and brain stem, it has also direct and indirect consequences on cortical activity. At the present time, there is no clear consensus on which EEG frequency bands or features best correlate with the NAN balance, but several EEG-derived NAN balance indices have been proposed. They include the composite variability index (CVI), the CI, and the CCS, the gradient between state and response spectral entropy (RE-SE), the qNOX, the gamma-band oscillations, and the A-line autoregressive index (AAI). In this chapter devoted to EEG-derived monitors, we will limit the description of NAN balance indices to those derived from the EEG.
The development of CVI was based on the fact that BIS™ value is influenced both by the cortical EEG and by the facial electromyogram. Although BIS ™ value before the occurrence of a noxious stimulation is not a reliable predictor of patient movement, its variability in response to this kind of stimulation is attenuated by antinociceptive medications. Facial EMG is also known to react to noxious stimulation when antinociception is not deep enough. CVI integrates the variability in BIS™ (sBIS) and in facial electromyogram (sEMG). Both sBIS and sEMG can be calculated from frontal EEG tracings. They correspond to the standard deviation of BIS and EMG over a 3 minute time period. The algorithm of CVI is based on a linear combination between sBIS and sEMG, and it was developed from a retrospective multicenter study dataset comprising 307 patients. The parameters entering the algorithm were chosen as a function of their ability to predict the occurrence of somatic events in response to noxious stimulation. The output is a dimensionless number between 0 and 100.
As mentioned The Cortical State and the Composite Cortical State , the CI is intended to quantify the magnitude of input to the cortex. It is proposed to assess the effect of antinociceptive medications or NAN balance. When combined with CCS, it seems to perform better than the combination of BIS ™ and CVI to predict response to stimulation, but it still needs further clinical validation.
Become a Clinical Tree membership for Full access and enjoy Unlimited articles
If you are a member. Log in here