Received Date: August 21, 2017; Accepted Date: September 19, 2017; Published Date: September 25, 2017
Citation: Hojong C, Tae-Ho Kim, Ki-duk K (2017) A Method for Quantifying the Depth of Anesthesia of Rats Based on Physiological Signal Model. J Bioengineer & Biomedical Sci 7: 236. doi:10.4172/2155-9538.1000236
Copyright: © 2017 Hojong C, et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution and reproduction in any medium, provided the original author and source are credited.
Visit for more related articles at Journal of Bioengineering & Biomedical Science
Measuring the depth of anesthesia using electroencephalogram (EEG) is an important and challenging task. Although various methods using EEG have been proposed, these algorithms quantify the depth of anesthesia without any physiological models. In this paper, a method to quantify the depth of anesthesia as well as a signal model that describes the changes in EEG during anesthesia is presented. The signal model is composed of numerous electrical signal sources and low-pass filters which model the microscopic signal from neurons and the electrical characteristics of the brain, respectively. Using the signal model, EEG is simulated as a summation of lowpass filtered impulse trains. The signal model suggests that the features of EEG change due to the decrease of the percent of pyramidal cells in the active state during anesthesia. Based on the signal model, an index for the depth of anesthesia, referred to as cortical activity index (CAI), is proposed. For the verification of the method, EEG signals from anesthetized rats are obtained and analyzed to evaluate the level of consciousness. The measurement results indicated that the proposed index, CAI, has achieves high stability over wide range of the depth of anesthesia. Moreover, it shows high correlation with other indexes such as modified detrend moving average (MDMA) and the WAVcns index. Based on the experimental results, CAI could be considered as a promising method to quantify depth of anesthesia with the plausible physiological signal model.
Depth of anesthesia; Anesthetic depth; Loss of consciousness; Level of consciousness; Anesthesia; Cortical activity
During surgery, it is very important to keep the patient at desired anesthetic depth. If the patient is not anesthetized enough, intraoperative awareness or postopera- tive recall can happen, possibly causing patient trauma. On the other hand, if the anesthetic agent is overused, the patient will require more time to recover from aftereffects. Preventing the overuse of anesthetics is especially important since anesthetics are powerful poisons. as the therapeutic index, a ratio of the dose that kill 50% of population to the effective dose for 50% of population, of general anes- thetics is only 3 − 4, whereas the indexes for many drug are on the order of several hundreds . Therefore, the anesthetic depth of the patient should be continuously monitored during surgery in order to use the proper amount of anesthetic agent. In recent years, a number of EEG-based anesthetic depth monitors have been developed based on the research results from both human and animal studies that changes in EEG during anesthesia reflect hypnotic state of the subject. Prior to others, the bispectral index (BIS) monitor from Aspect Medical Systems showed very successful results in 1997 . Since then, other algorithms for measuring anes- thetic depth have been proposed and commercialized. For example, the Narcotrend monitor  from MonitorTechnik, Entropy monitor  from Datex Ohmeda, and NeuroSENSE monitor  from NeuroWave Systems were developed and received considerable attention. Although there have been several successful results, still many researchers are concentrating on developing an anesthetic depth monitor. The reason is that those monitors sometimes fail under certain circumstances de- pending on the type of anesthetic agent or target person . One of the main reasons for the failures is the lack of a signal model that quantitatively describes the change in EEG during anesthesia in a unified way. Due to the absence of a sig- nal model, the intrinsic and quantitative relationships between EEG and the depth of anesthesia have not been thoroughly understood. Thus, previous approaches to estimate the depth of anesthesia from the EEG signal sought mathematically welldefined parameters, such as bicoherence , cepstral analysis , entropy [8-11], self-affinity [1,6], and wavelet coefficients [12,13], which seem to be highly correlated with the clinically evaluated anesthetic depth.
In this work, a signal model that gives a quantitative description of the changes in EEG with respect to the change of the percent of pyramidal cells in the active state (PA) is presented. The signal model can explain not only a lot of features shown in EEG during anesthesia such as burst suppression and the decrease in high frequency components but also the changes in the indexes for the depth of anesthesia. Moreover, based on the signal model, a time-domain method is proposed in order to estimate the depth of anesthesia from EEG segments. In order to verify the proposed algorithm, EEG signals from anesthetized rats are measured, and the depth of anesthesia is estimated using the method. Not only does the result from the proposed method show high correlation with other indexes for the depth of anesthesia, but also it outperforms in terms stability over a wide range of anesthetic depths.
In order to establish the signal model of EEG, the mechanism of the generation of EEG should be understood. The main sources of EEG are the local potentials caused by extracellular current sinks and sources (LPCSs) from the pyramidal cells in the cortex layer [14,15]. These signals summate and are recorded as EEG activity at the surface of the skin. In other words, EEG activity represents the summated electrical activity of multiple sources located in the cortex layer, with the most significant contribution coming from the sources located in superficial cortical layers.
Studies on the cellular pharmacology of anesthetic agents have shown that anesthetics hyperpolarize neurons by increasing inhibition or decreasing excitation. During wakefulness, excitatory input from other neurons provides a depolarizing drive that causes neurons to exhibit single-spike tonic firing. During anesthesia, neurons switch into burst-firing mode. At intermediate anesthetic concentrations, neurons oscillate between an active up-state and an inactive downstate. As anes- thetic doses increase, the up-state turns to a short burst and the down-state be- comes progressively longer . Therefore, the percent of pyramidal cells in the active state PA can be considered a parameter that directly reflects the depth of anesthesia.
Before going into detail on the signal model, it should be noted that the purpose of the signal model is to explain the changes in the features of EEG during anesthesia, but not to solve general neuroelectric inverse problems. Based on the physiological bases discussed in the previous section, the signal model is established with the following statements.
• EEG is generated by superimposing the LPCSs from many signal sources in the cortex layer.
• LPCSs are low-pass filtered due to the impedance of the channel before su- perposition and the frequency response of the filters differs depending on the location of neurons.
• The percent of pyramidal cells in the active state PA is proportional to the level of consciousness.
In the proposed signal model, the LPCSs generated from signal sources are low- pass filtered and added as shown in Figure 1. Each source generates LPCSs when it is in the active up state and becomes silent when it is in the inactive down state. The sources are grouped into three classes depending on the frequency response of the low-pass filters. The filters with high cut-off frequency are used to model the superposition of the signals from the sources located near the electrodes and vice versa. Not only the cut-off frequencies but also the low frequency gains of the filters are adjusted to reflect the signal attenuation due to the resistance of the brain. Some sources are modelled to generate positive LPCSs while others generate negative LPCSs .
Figure 1: The signal model of EEG with respect to the depth of anesthesia. Numerous LPCSs are superimposed after low-pass filtering due to the impedance of the brain. The signal sources are modeled to generate tones when they are in the active up state and to be silent during the inactive down state. The percent of pyramidal cells in the active state, PA, is assumed to be proportional to the level of consciousness.
Figure 2 shows sample EEG signals generated with the proposed signal model using MATLAB. The specifications of the simulator are summarized in Table 1. It can be seen that the change in the waveforms with respect to the change of PA is highly consistent with the change in the EEG waveform with respect to the change of the depth of anesthesia. When PA=40%, a decrease in high frequency components can be seen compared to the case when PA=100%. When PA=10% and PA = 100%, burst suppression, a pattern that appears during deep anesthesia, is observed.
|Sampling frequency||128 Hz|
|Type of LPFs||Maximally flat (IIR)|
|Cutoff frequency ofLPFs||1Hz (HIP), 10Hz (MIP), 40Hz(LIP)|
|Number of signalsources||50 (HIP), 10 (MIP), 5(LIP)|
|Low frequency gain||0.1 (HIP), 0.5 (MIP), 1(LIP)|
|Time interval between LPCSs from a single source||20ms in average (up state)
∞ (down state)
Table 1: Specification of EEG generator (HIP: high impedance path, MIP: medium impedance path,LIP:lowimpedance path).
In order to verify whether the proposed signal model substantially reflects the change in the feature of EEG during anesthesia, generated EEG signals are evaluated in terms of several indexes which are known to be highly correlated to the depth of anesthesia. The exponent from modified detrended moving average (MDMA) analysis , the WAVCNS index  and spectral entropy EntropyCA are well-known parameters which have high correlation with the depth of anesthesia.These parameters are evaluated for generated EEG signals with various PAs as shown in Figure 3. The relationship between PA and CAI, which is the index for anesthetic depth derived in this work, is also presented in the figure. Gaussian distribution function is used as the probability distribution function of wavelet coefficients required for the calculation of the WAVCNS index, in order to avoid high dependence on training data. It can be seen that all parameters are highly correlated with the percent of pyramidal cells in the active state.
Figure 3: (a) The exponent from modified detrended moving average analysis (MDMA), (b) specral entropy (SpEn), (c) the WAVCNS index and (d) the cortical activity index (CAI) are calculated for generated EEG segments. The length of each EEG segment is 16384 samples, which corresponds to 128 s in time.
According to the signal model, measuring the depth of anesthesia from an EEG segment can be substituted by estimating the number of LPCSs, which are super- imposed on the EEG segment, since the number of LPCSs is proportional to the percent of pyramidal cells in the active state. As it is almost impossible to actually distinguish each LPCS from EEG, we propose a simple time-domain method to estimate the density of LPCSs. First, the signal is high-pass filtered, and both the high-pass filtered signal and unfiltered signal are cut into epochs using a sliding window. Each epoch from the high-pass filtered signal is normalized by the RMS value of the corresponding epoch from the unfiltered signal, and the points above a threshold value in the high-pass filtered epoch are counted. Subsequently, an index is obtained by dividing the number of points above the threshold value by the total number of points in the epoch. The resulting index is inherently bounded between 0 and 1. This method for measuring the depth of anesthesia can be intuitively un- derstood as follows. Dynamic EEG epochs from conscious subjects will still contain a considerable number of points above the threshold after high-pass filtering. A large portion of slow EEG epochs from sedated subjects will be suppressed due to high-pass filtering, leaving a small number of points above the threshold value. When the input epoch is isoelectric, the value of the index is clearly 0.
One problem that may occur due to normalizing each epoch by RMS value is the amplification of noise in isoelectric epochs. Since the signal power of the EEG component in an isoelectric epoch is low, noise in an isoelectric epoch is inevitably amplified. This may mislead to a wrong anesthetic depth as noise and EEG from conscious subjects often show similar characteristics, especially in time domain. In order to compensate for this effect due to normalization, a dynamic threshold decision method is applied. The concept of a dynamic threshold decision is to increase the threshold value when the signal power of EEG component is low. The threshold for each epoch is obtained by calculating the signal power in the low frequency band as :
and Kdenote an unfiltered normalized epoch, an index that corresponds to 4 Hz in discrete-time domain and an experimentally-determined coefficient, respectively.
In the experiment, eight rats (Sprague Dawley), four males and four females, were involved. The animals were six to ten weeks old and weighted between 150 and 250 g. The animals were treated under environmentally controlled conditions (12 h light/dark cycles, 19−23°C), with food and water ad libitum. Institutional Animal Care and Use Committee of KAIST approved the experimental protocol used in this study.
The environmental setup for the measurement of rat EEG during anesthesia is shown in Figure 4. Before the measurement, the scalp hair of the rats was shaved while the rats were anesthetized with 2.5-3.5 mL/kg mixed anesthetic agent (volumetric ratio 4:1 of zoletil and rompun, respectively) administered intraperitoneally. All rats had three to five days of relaxation period after shaving. A BIS VISTA mon- itor (Version 3.00, Aspect Medical Systems), which captures signals at 128 Hz in 0.05 μV resolution, was employed for the measurement of EEG. Copper alligator clips were used as the electrodes for measuring EEG, as the BIS electrodes are designed to be used for human. Signal electrode, the reference electrode and the ground electrode were clipped to the middle of scalp, the left ear and the right ear, respectively. Before the electrode placement, the fur on the forehead of the rat was shaved and pasted with conductive gel to ensure electrical conductivity. The measurements were conducted after attaching the electrodes to the rats after initial anesthesia, and the rats were kept in a chamber during the measurement. In order to avoid sudden deep anesthetization and sudden emergence, the gas was delivered to the chamber with a constant flow rate for the continuous change in the concentration of anesthetic instead of directly delivering the gas from the anes- thetic machine to the rat. Artificial respirator in the anesthesia machine was not used since the tidal volume of the respirator is not appropriate for rats. Enflurane, which is widely used for anesthetizing animals, was used as the anesthetic agent. During induction, anesthetic gas with a flow rate of 0.5 L/min and a concentration of Vol. 5% was delivered to the chamber. The gas delivery was stopped for 5 to 10 minutes for maintenance when the rat was sufficiently anesthetized. After the maintenance phase, oxygen was delivered into the chamber to induce emergence from anesthesia with a flow rate of 0.4 L/min. Measurements were concluded when the rat woke from the anesthesia.
After transferring the raw EEG signal obtained from the BIS VISTA monitor to a computer, the data is processed using MATLAB. First, the raw EEG is filtered by an FIR filter with a zero at 60 Hz for power supply noise reduction and linear phase response for the prevention of waveform distortion. Second, the low-pass filtered signal is filtered again by an FIR high-pass filter with the 3 dB frequency of 33 Hz. Both low-pass filtered and high-pass filtered signals are cut into 16s epochs, where the high-pass filtered epochs are used to count the number of samples over the threshold and the low-pass filtered epochs are used for the calculation of threshold value and RMS value. Before calculating the threshold value and RMS value from the low-pass filtered epochs, low frequency noise and artifacts are removed using a denoising technique based on wavelet decomposition . Low frequency artifacts are removed by bounding wavelet coefficients in the lower frequency band without distorting the waveform. This denoising is particularly important since, without it, the threshold value can be seriously affected by artifacts.
For most of the measurement, the anesthetic depth calculated and exported from the BIS monitor was invalid. The difference between humans and rats and the low signal quality due to the modification of electrodes are probably the main reasons. Therefore, in this work, the scaling exponent from the MDMA method and the WAVCNS index, which are the most up-to-date measures of the depth of anesthesia are used as the references for the depth of anesthesia.
The anesthetic depths are evaluated for eight measured EEG signals, and the results are shown in Figure 5. The scaling exponent from the MDMA method is reversed in the figures for easier comparison. Postprocessing, such as filtering the indexes, is not applied in order to observe the variability and response time. It can be seen that the result from the proposed method has high consistency with the results from other methods and low variability over a wide range of anesthetic depth. The Pearson correlation coefficients between CAI and the exponent from the MDMA method and that between CAI and WAVCNS , calculated from all eight data, are −0.9365 and 0.9049, respectively.
In this work, a method for quantifying the depth of anesthesia is presented. Un- like other recent works that rely on advanced signal processing techniques, the proposed method has a simple and intuitive algorithm based on a physiological signal model that suggests the intrinsic relationship between the depth of anesthesia and the waveform of EEG. The experimental results show that the proposed algorithm is an attractive method for quantifying the depth of anesthesia in terms of both complexity and stability over a wide range of anesthetic depths. With the aid of the signal model, which suggests that evaluation of the depth of anesthesia can be performed by estimating the number of LPCSs superimposed on EEG, it might be possible to develop more advanced method for quantifying the depth of anesthesia without clinical experiments. Although the measurement result in this work is limited to the EEG from rats, the authors expect that this result can be easily extended to human applications.
The authors gratefully acknowledge the partial financial support pro- vided by KI institution-specific project of 2017 KAIST’s own research projects and Industry research and development project (G01160221).