Oral surgery Local anesthesia Oral surgery Local anesthesia.
Review: Intelligent Modeling and Control in Anesthesia · Journal of Medical and Biological...
Transcript of Review: Intelligent Modeling and Control in Anesthesia · Journal of Medical and Biological...
Journal of Medical and Biological Engineering, 32(5): 293-308 293
Review: Intelligent Modeling and Control in Anesthesia
Jheng-Yan Lan1,2 Maysam F. Abbod3 Rong-Guan Yeh2
Shou-Zen Fan4 Jiann-Shing Shieh2,5,*
1Department of Anesthesiology, National Taiwan University Hospital in Yuan Lin Branch, Yuanlin 640, Taiwan, ROC 2Department of Mechanical Engineering, Yuan Ze University, Chungli 320, Taiwan, ROC
3School of Engineering and Design, Brunel University, London UB8 3PH, United Kingdom 4Department of Anesthesiology, College of Medicine, National Taiwan University, Taipei 106, Taiwan, ROC
5Center for Dynamical Biomarkers and Translational Medicine, National Central University, Chungli 320, Taiwan, ROC
Received 5 Sep 2011; Accepted 17 Feb 2012; doi: 10.5405/jmbe.1014
Abstract
This paper provides a detailed review of the clinical aspect and engineering view of how to measure, interpret,
model, and control general anesthesia. The mechanisms of anesthesia in terms of unconsciousness, amnesia, analgesia,
and akinesia in modern balanced anesthesia are reviewed. The assessment and interpretation of anesthesia according to
clinical signs (i.e., lacrimation, sweating, papillary dilatation), physiological monitors (i.e., electromyography,
electrocardiography, blood pressure, electroencephalography, oxyhemoglobin saturation), and evaluation indices (i.e.,
bispectral index scale, entropy, auditory evoked potentials, and surgical stress index) are reviewed in order to define the
objectives of general anesthesia. Finally, the intelligent modeling and control of anesthesia are thoroughly reviewed.
Modern general anesthesia is moving towards the monitoring, interpretation, modeling, and control of multi-inputs
from quantitative and qualitative nonlinear physiological signals and multi-outputs for drug control of unconsciousness,
amnesia, analgesia, and akinesia. A multistage hierarchical system should thus be developed for the modeling and
control of general anesthesia in the future.
Keywords: General anesthesia, Unconsciousness, Amnesia, Analgesia, Akinesia, Intelligent systems, Modeling and
control
1. Introduction
General anesthesia is a reversible change that includes
unconsciousness, amnesia, analgesia, and akinesia, with
concomitant stability of the cardiorespiratory and autonomic
systems [1-5]. In order to reach the proper anesthetic state for
surgery, hypnotic and analgesic effects are considered as major
fundamental pharmacologic components. Under general
anesthesia, memory and awareness are critical components of
the depth of anesthesia (DOA). However, anesthesiologists
have multiple inconsistent definitions of the anesthetic state
and have no standard measurement to assess it. Analgesia is an
important aspect of balanced anesthesia, but there is no
commercially available monitor for its evaluation. Therefore, in
clinical practice, anesthesiologists must provide specific care
during surgical procedures, including neuromuscular relaxation
or paralysis, adequate DOA, and analgesia. These interacting
requirements necessitate a balanced administration of suitable
drugs.
* Corresponding author: Jiann-Shing Shieh
Tel: +886-3-4638800 ext. 2470; Fax: +886-3-4558013
E-mail: [email protected]
Direct measurements of blood pressure (BP), respiratory
parameters, temperature, blood oxygen saturation, and other
related vital signs are feasible. The patient can be controlled by
manipulating the monitored values, but the response is often
delayed. For example, when tachycardia and hypotension
occur, the total amount of blood loss is large and the
hypoperfusion continues for a period of time. Furthermore,
direct measurements cannot provide sufficient information of
the autonomic nervous system (ANS) and central nervous
system (CNS), which are related to the DOA, level of surgical
stress, and nociceptive changes. Biological systems are
spatially and temporally complex systems connected by
dynamic interconnected feedback loops. Thus, although direct
measurements of vital signs are available, precise modeling and
anesthesia control are very challenging for anesthesiologists.
Quantitative modeling approaches are being replaced by
qualitative techniques, especially in the artificial intelligence
field. Hybrid models (quantitative/qualitative) and hybrid
intelligent algorithms (neural networks, fuzzy logic,
evolutionary computing) have been applied to anesthesia [6].
This review is divided into two sections. The first section
reviews the mechanisms and effects of anesthetics and how to
assess the anesthesia. The second section reviews intelligent
modeling and control systems for anesthesia.
J. Med. Biol. Eng., Vol. 32 No. 5 2012 294
2. Mechanisms of anesthesia
The most commonly used inhaled anesthetics in modern
anesthesia are a single gas (i.e., nitrous oxide) and volatile
liquids (i.e., halothane, enflurane, isoflurane, desflurane, and
sevoflurane). The immobilizing and hypnosis/amnesia effect of
inhaled anesthetics involves sites of action in the spinal cord and
supraspinal mechanism, respectively. Each agent has a
complicated molecular mechanism and multiple targets
contribute to the effects. Nonopioid intravenous anesthetics such
as barbiturates (i.e., thiopental, methohexital), benzodiazepines
(i.e., midazolam), propofol, ketamine, and etomidate, have an
important role in modern anesthesia practice. They are widely
used to facilitate the rapid induction of anesthesia or to provide
sedation during monitored anesthesia care. With the introduction
of propofol, intravenous anesthesia has become more popular as
a component of anesthesia maintenance. However, currently
available intravenous anesthetics are not ideal anesthetic drugs
in the sense of producing all and only desired effects (hypnosis,
amnesia, analgesia, immobility). Therefore, balanced anesthesia
with multiple drugs (neuromuscular-blocking drugs, inhaled
anesthetics, and opioids) is generally used.
2.1 Mechanisms of unconsciousness
Most anesthetics induce unconsciousness by changing
neurotransmission, either increasing the primary inhibitory
neurotransmitter gamma-aminobutyric acid (GABA) or
decreasing the activation of the primary excitatory N-methyl-
D-aspartate (NMDA) receptors [7,8] at multiple sites in the
cerebral cortex [9-11], thalamus, and brain stem [12]. A
positron-emission tomographic study showed that cortical
metabolic activity is decreased after general anesthesia [13],
and functional magnetic resonance imaging has shown
evidence of unconsciousness in the cerebral cortex [10].
Hypnotic drugs administered during induction rapidly reach
brain-stem arousal centers and contribute to unconsciousness.
Furthermore, the drugs act on several nuclei in the area,
resulting in impaired brain-stem function which is manifested
as apnea [14], atonia [15], and loss of normal reflexes, such as
oculocephalic and corneal reflexes [16]. In the central
thalamus, anesthetics control the normal arousal system [17].
Most anesthetics interact with various molecular sites of
neuronal action via ligand-gated or voltage-gated channels on
the main targets and cause hyperpolarization of neurons in
thalamocortical loops, disrupting the connectivity in the cortex
[18]. In the mammalian CNS, GABA is the main inhibitory
neurotransmitter and its fast inhibitory effects are controlled
through GABAA receptors. Barbiturates, etomidate, propofol,
sevoflurane, enflurane, desflurane, and isoflurane are GABA-
agonist agents [9], which increase the sensitivity of receptors to
GABA, prolong the inhibitory post-synaptic effect after a
GABA release, and increase the inhibition of postsynaptic
neuronal excitability [18] at clinically effective concentrations.
Glutamate is the major excitatory neurotransmitter in the
CNS. Ketamine, nitrous oxide, and xenon can inhibit glutamate
effects through the NMDA receptor [18-20]. Most hypnotic
agents produce low electroencephalography (EEG) patterns
under unconsciousness conditions, whereas NMDA antagonists,
such as ketamine, are associated with active EEG patterns.
When ketamine is used with propofol, an additive anesthetic
interaction occurs that is not reflected in the bispectral index
scale (BIS) [21]. Thus, the addition of ketamine to an anesthetic
complicates the interpretation of the processed EEG, again
emphasizing the need to consider the clinical circumstances
and drugs used when interpreting a DOA index.
2.2 Mechanisms of analgesia induced by general anesthesia
Intravenous or inhalational anesthetics are used for
hypnotic effects, but are ineffective at eliminating the
hemodynamic response to high-stress stimuli even with large
doses [22,23]. Opioids are typically combined with hyponotic
drugs to create a state of non-responsiveness. Opioids are unique
in producing analgesia without loss of touch, proprioception, or
consciousness. Opioids suppress pain by their action in the
brain, spinal cord, and peripheral nervous system through
agonists, partial agonists, and mixed agonists- antagonists, with
opioid receptors (μ-,δ-, κ-opioid receptors). The principal effect
of opioid receptor activation is a decrease in neurotransmission,
which results mostly from presynaptic inhibition of
neurotransmitter release (e.g., acetylcholine, dopamine,
norepnephrine, substance P) [24] and increased potassium
conductance, leading to hyperpolarization of cellular
membranes. Opioids produce analgesic effects by directly
inhibiting the ascending transmission of nociceptive stimulation
from the spinal cord dorsal horn and activate pain control
circuits via the descending transmission of the rostral
ventromedial medulla.
Opioids affect many organ systems, including the
respiratory and cardiovascular systems, and can cause a variety
of adverse effects. The pharmacokinetic and pharmacodynamic
properties of opioids are affected by a variety of factors, such
as age, body weight, organ failure, and shock. Proper dosing
and monitoring allow the adverse effects to be minimized.
Increasing the concentration of inhaled anesthetics produces a
continuum of changes in the EEG and eventually results in
burst suppression and a flat EEG. In contrast, a ceiling is
reached with opioids. Once the opioid dosage is increased to a
critical level, further increases will not affect the EEG [25].
Problems with signal processing need to be resolved before
analysis of the EEG can be used as a routine parameter of
DOA.
2.3 Mechanisms of akinesia induced by general anesthesia
Neuromuscular blockers are used to facilitate endotracheal
intubation and maintain neuromuscular blockade during
surgical procedures. Neuromuscular-blocking drugs interrupt
the transmission of nerve impulses at the neuromuscular
junction and thereby produce paresis or paralysis of skeletal
muscles. Non-depolarizing muscle relaxants produce a
neuromuscular blockade by competing with acetylcholine for
postsynaptic α-subunits. In contrast, succinylcholine, a
depolarizing neuromuscular blocker, produces prolonged
depolarization that results in decreased sensitivity of the
Intelligent Modeling and Control in Anesthesia 295
Table 1. Clinical sign features, power spectra, and entropy analysis evaluation indices for inadequate general anesthesia conditions (EMG:
electromyography; EEG: electroencephalography; ECG: electrocardiography; SpO2: oxyhemoglobin saturation; BP: blood pressure; HRV:
heart rate variability; RRI: R peak interval; FFT: fast Fourier transform; LFP: low-frequency power; HFP: high-frequency power; PPGA:
plethysmograph amplitude; BPV: blood pressure variability; BIS: bispectral index; AEP: auditory evoked potentials; CPC: cardiopulmonary
coupling; LHR: low/high frequency ratio; DFA: detrended fluctuation algorithm; MSE: multi-scale entropy; CI: complex index; SSI: surgical
stress index; and PVI: pleth variability index).
Clinical signs of inadequate general
anesthesia
Monitors Signal
processing
Evaluation
indices
Somatic responses
Motor Movement
Withdrawal
EMG,
Observation;
Power spectral
analysis, BIS, Entropy
BIS,
Entropy AEP Consciousness
Nociception
Awareness,
Pain
EEG,
Subjective
experience;
Autonomic responses
Breathing Breathing pattern
change
Observation;
Respiration rate, volume
Coherence and
Cross-power Spectrum
CPC
(combined with ECG)
Hemodynamic Tachycardia ECG HRV, RRI, FFT,
LFP, HFP
LHR,DFA
(α index), MSE (CI),
SSI, PVI Volume SpO2 PPGA,
Hypertension BP BPV
Sudomotor Sweating Skin
conductivity
Papillary
dilatation
Pupillometry
Lacrimation Observation
Hormonal Catecholamines,
Corticosteroids
Blood drawing and
lab analysis
postsynaptic nicotinic acetylcholine receptor and inactivation
of sodium channels, inhibiting propagation of the action
potential across the muscle membrane.
Doctors often use a peripheral nerve stimulator to titrate
the dose to produce the desired pharmacologic effect. At the
conclusion of surgery, the responses evoked by the nerve
stimulator are used to judge spontaneous recovery from
neuromuscular blockade, which is facilitated by the
administration of anticholinesterase drugs. Muscle relaxants
may confound the accuracy of devices in relation to the DOA.
Fluctuating muscle tension and tonic input from muscle
spindles contribute to CNS arousal. As a result, muscle
relaxants may deepen anesthesia [26-29] by reducing neural
traffic [30]. The reversal of muscle relaxation can also change
the amount of electromyography (EMG) activity and influence
the anesthesia monitor [31].
3. Assessment and interpretation of anesthesia
Inadequate general anesthesia and pain control may cause
side effects and lead to increased morbidity. Although the
assessment of awareness and pain is always based on patients’
subjective reports, there are clinical signs (i.e., lacrimation,
sweating, papillary dilatation), physiological monitors (i.e.,
EMG, electrocardiography (ECG), BP, EEG, oxyhemoglobin
saturation (SpO2)), and evaluation indices (i.e., BIS, entropy,
auditory evoked potential (AEP), surgical stress index (SSI))
that can help give objective reports of general anesthesia.
Table 1 shows clinical sign features and their correlation to
power spectral and entropy analysis evaluation indices (e.g.,
BIS and AEP) for inadequate general anesthesia conditions.
The assessment and interpretation of general anesthesia are
introduced in the following subsections.
3.1 Clinical signs
There are three periods of general anesthesia: induction,
maintenance, and recovery. Different clinical signs can be
observed in the individual stages. When a small dose of a
hypnotic drug such as propofol or a barbiturate is administered,
the patient is in the sedation state and easily arousable. As the
dose is increased, a state of paradoxical excitation is observed,
which is characterized by incoherent speech or purposeless
movements. As the concentration is further increased, an
irregular respiratory pattern develops, which may progress to
apnea. At the same time, muscle tone and response to oral
commands obviously decrease. The corneal and eyelash
reflexes are lost. Tracheal intubation is usually performed at
this state.
A combination of inhalational agents, hypnotic agents,
opioids, and muscle relaxants is used to maintain anesthesia. In
the maintenance period, anesthesiologists use heart rate (HR)
and BP to monitor the level of anesthesia because the clinical
signs are difficult to access. When the HR and BP increase
dramatically, it means that the state of anesthesia is inadequate
for the patient. Other clinical signs that indicate inadequate
general anesthesia are sweating, papillary dilatation, the return
of muscle tone, movement, lacrimation [32], and the hormonal
response [33,34] (i.e., catecholamines, corticosteroids). PRST
(blood pressure, heart rate, sweat, and tear formation) is used to
assess responsiveness. It is used as a standard measure of
autonomic reactions in clinical practice [32].
The return of spontaneous respiration and muscle tone,
J. Med. Biol. Eng., Vol. 32 No. 5 2012 296
HR and BP increases, salivation and tearing, and grimacing,
swallowing, coughing, and defensive movements are the
behavioral and physiological signs of recovery from general
anesthesia. At this point, there is sufficient return of brain-stem
reflexes to maintain spontaneous respiration and airway
protection, and thus the anesthesiologist performs extubation.
3.2 Electrophysiologic monitoring
The EEG represents electrical activity of the cerebral
cortex derived from summated inhibitory and excitatory
post-synaptic activity, which in turn is the result of subcortical
events, including those that pass through subcortical thalamic
nuclei. Anesthetic drugs affect cerebral blood flow, metabolism,
and EEG patterns, which are related to the degree of EEG
activity [30]. For an awake state with closed eyes, EEG is
active and normally shows prominent alpha activity (10 Hz).
After the administration of a hypnotic drug, the patient enters
the sedation state and the EEG shows an increase in beta
activity (13-25 Hz) [5]. In the maintenance period, there are
four phases of EEG. Phase 1 is a light state of general
anesthesia with decreased EEG beta activity (13-30 Hz), and
increased alpha activity (8-12 Hz) and delta activity (0-4 Hz)
[31]. In phase 2, an intermediate state, beta activity decreases
and alpha and delta activities increase [31,32]. During phase 3,
a deeper state, the EEG is characterized by burst suppression
[33]. Surgery is often performed in phases 2 and 3. Phase 4 is
the most profound state, during which the EEG is isoelectric.
EEG patterns were initially used as a measure of the DOA
to define the anesthetic state but a discrepancy seemed to exist
between the clinical signs and EEG patterns, especially during
induction and recovery [35]. Because the effects of the
interactions of several administered anesthetic drugs on EEG
patterns are not known, there is no standard method for
choosing an optimal EEG parameter, and there is no clear
definition for the assessment of clinical DOA, it is difficult to
apply EEG to measure clinical DOA.
3.3 Quantitative EEG-derived indices and monitors
Power spectral analysis, a frequency analysis method that
decomposes an EEG signal into a series of sine waves, is a
traditional method for the quantification of EEG and reflects the
amount of activity in the frequency bands of the EEG signal.
Bispectral analysis was applied to EEG signals to describe the
states of awake and sleeping subjects by Barnett et al. [36] and
Dumermuth et al. [37]. Numerous methods for processing EEG
signals, such as entropy, the patient state index, narcotrend, and
evoked responses, have been evaluated for monitoring DOA.
3.3.1 Bispectral index scale
The BIS value is calculated from a multivariate logistic
regression analysis from a collected database of EEG recordings
from adult volunteers with hypnotic drug concentrations,
behavioral assessments, and clinically important endpoints
[38,39]. The BIS, scaled from 0 to 100, creates a dimensionless
number. Zero represents complete electrical silence of the
cerebral cortex, and 100 represents an alert state. A BIS value of
less than 55 is acceptable during general anesthesia, and that
between 50-60 is associated with low probability of response to
verbal command.
In a prospective cohort study [40], 4945 surgical patients
monitored with the BIS were compared with 7826 similar cases
in a previous study where no cerebral monitoring was used. In
the BIS-monitored group, 0.04% had explicit recall compared
with 0.18% in the control group (p < 0.038). BIS values
between 40 and 60 have been recommended. Hence, this cohort
study and similarly other studies have shown that the incidence
of awareness decreasing from 0.18% to 0.04% is associated with
the use of BIS monitoring [41,42]. However, in some studies,
there was no difference in the incidence of awareness and
volatile anesthetic consumption between BIS-monitored and
routine care patients [43]. In addition, the BIS has been applied
to pediatric anesthesia [44,45], the detection of brain death [46],
the prediction of neurological prognosis after trauma and
cerebral ischemia [47], and as a monitor of the sedation level in
intensive care units [48].
The BIS is used as a DOA monitor during propofol
anesthesia, but it has some limitations for sevolfurane and
enflurane anesthesia. During ketamine, dexmedetomidine, N2O,
and xenon anesthesia, the BIS does not perform well [49].
3.3.2 Entropy
Approximate entropy, which measures the logarithmic
likelihood of patterns, has been used to analyze physiological
signals and EEG for short time series [50-53]. Although there
are multiple entropy algorithms, only spectral entropy has been
used in a commercially available device, the GE Healthcare
Entropy Module (formerly Datex-Ohmeda M-Entropy), which
was commercially released in 2003. The algorithm uses the
time-domain and frequency-domain data from Fourier analysis
and then applies the Shannon function to the frequency data.
Spectral entropy, termed time-frequency balanced spectral
entropy, depends on the sampling frequency and windowing
[54], but is independent of the frequency components of the
signal [55]. Both approximate and spectral entropy decrease
during general anesthesia [54].
Facial EMG has a frequency range that overlaps with
traditional EEG (0.8-32 Hz), thereby confounding the analysis
of cortical activity. Facial EMG changes with the level of
consciousness and the use of neuromuscular-blocking drugs.
When original EEG signals are analyzed, the response entropy
(RE) and state entropy (SE) indices are obtained. SE is
computed from the EEG across the 0.8 to 32 Hz range and
should encompass mainly the hypnotic elements of the EEG,
whereas RE is computed from 0.8 to 47 Hz, which includes a
significant amount of facial EMG [56]. The displayed SE range
is 0 (isoelectric EEG) to 91 (fully awake), and the RE range is
0 to 100. The anesthetic range is 40 to 60. The manufacturer
recommends that SE values outside this range may require a
change in hyponotic dosing. If the SE value is in this range but
the RE is more than 10 above the SE, more analgesic may be
required.
3.3.3 Evoked potentials
Sensory, including visual, somatosensory, and auditory,
stimulation produces an evoked response between the sensory
Intelligent Modeling and Control in Anesthesia 297
receptor and neural generator within the functional integrity
pathway in the CNS. The evoked response can be separated
from the underlying, spontaneous EEG by signal-averaging
techniques. Because auditory evoked responses are sensitive to
anesthetic drugs and their signal represents the conduction of
the peripheral nerve to the brain stem, which is the major
control system of the ANS, AEP has been used as a measure of
the anesthetic effect and DOA [57].
Many investigations associated with evoked potentials
have focused on mid-latency auditory evoked potentials
(MLAEPs) [57-59]. Hypnotic anesthetic drugs increase the
latencies and decrease the amplitudes of waves Pa and Nb. In
contrast, opioids produce minimal changes in MLAEPs [60] in a
clinically effective concentration. There are some limitations of
MLAEPs, such as complex setup, need for intact hearing, and
considerable time needed to produce a response. Therefore, the
AEP index, based on a proprietary algorithm, was introduced
[61,62]. The AEP index simplifies the interpretation of MLAEP
waveforms, but it still requires an adaptive technique for
extracting the MLAEP from the original EEG signal. The
process involves an autoregressive method and an exogenous
input (ARX) to allow the extraction of the AEP signal within
15-25 sweeps of 110-ms duration [63]. This concept has been
incorporated into a commercial device that calculates the A-Line
ARX Index (AAI), which was derived from the fast-extracted
MLAEP waveform analysis and ranges from 0 (deep hypnotic
effect) to 100 (awake). The recommended surgical anesthesia
range is 15-25. In clinical studies, the A-Line performed
comparably to the BIS and other EEG-based monitors but did
not demonstrate advantages that would outweigh the nuisance of
ear plugs and continuous high-frequency clicking sounds in
patients’ ears [64-66].
3.3.4 Clinical interpretation and effects of drug effects in EEG
Studies of all currently available devices show a good
relationship with the Modified Observer’s Assessment of
Alertness/Sedation (MOAA/S) scale for commonly used
hypnotics such as midazolam [67], propofol [66,68-72],
isoflurane, sevoflurane [64,73], and desflurane. The relationship
between sedation and the derived electrophysiologic index does
not seem to be influenced by age in adults [74]. Although the
average index tracks the sedation scale well at the steady state,
there is huge inter-individual variation, particularly during
non-steady-state conditions such as induction and recovery [75].
Thus, inter-subject variability in the EEG requires the
assessment of the awake values before titration of a hypnotic to
a given target EEG index.
The dissociative anesthetic ketamine, an NMDA antagonist,
produces excitatory effects in the EEG. Ketamine doses that
create unresponsiveness do not change the BIS [76]. When
ketamine is used with propofol, an additive anesthetic
interaction occurs [77] that is not reflected in the BIS [21]. Thus,
the addition of ketamine to an anesthetic complicates the
interpretation of the processed EEG.
Nitrous oxide (N2O), an NMDA antagonist, has a
sympatholytic effect. Most studies suggest that nitrous oxide
given in concentrations of up to 70% has minimal effect on the
BIS [78,79] and have found that the effect of nitrous oxide
appears to depend on the monitor used.
3.4 Neuromuscular monitoring
Neuromuscular function is monitored by evaluating the
muscular response to supramaximal stimulation of a peripheral
motor nerve [80,81]. After the administration of a
neuromuscular-blocking drug, the decrease of muscle response
reflects the degree of fiber blockade [82,83]. There are several
patterns of nerve stimulation. The most commonly used patterns
of electrical nerve stimulation are single-twitch, train of four
(TOF), titanic count, post-tetanic count (PTC), and double-burst
stimulation (DBS). Moreover, it has been possible to measure
movement (i.e., degree of muscle relaxation) since the
commercial product the Datex EMG machine (Datex NMT-221
monitor, Datex Instrumentarium, Helsinki, Finland) launched in
the market in 1980. Therefore, it is easy to monitor
neuromuscular function in general anesthesia.
3.5 Measurements of analgesia during general anesthesia
Pain is a subjective conscious experience derived from
internal or external stimuli. During general anesthesia,
consciousness vanishes and the experience of pain disappears.
The term nociception is used to describe the consequences of
surgical stimulus on system function. Several indicators,
including heart rate variability (HRV) [32], changes of the
photoplethysmographic pulse wave amplitude [84,85], facial
muscle activity [86], changes in skin conductivity [87],
pupillometry [88], ocular microtremors [89], and EEG-based
monitoring [86], have been proposed to assess the degree of
nociception.
Photoplethysmography (PPG) gives the concentration of
oxyhemoglobin in the blood and its waveform can be used to
estimate volume changes. Changes in PPG amplitude (PPGA)
correlate with perfusion and reflect the relation of the left
ventricular volume and the peripheral vascular bed [90,91].
Under general anesthesia, increased PPGA response [92] and
skin vasomotor response [93] have been related to nociception.
However, several factors may easily cause noise in PPGA
measurements, such as the movement of the detector,
physiological factors, and pharmacological factors.
Huiku et al. [94] proposed the SSI, which is a simple
numerical measure of the surgical stress level under anesthesia.
Two continuous variables, the interval between successive
hearts beats (HBI) and PPGA, are used to calculate SSI: 100-
(0.7 PPGAnorm + 0.3 HBInorm), where PPGAnorm is the
normalized PPGA and HBInorm is the normalized HBI. The
SSI value indicates the balance between the intensity of
surgical stimulus and the level of antinociception. An SSI value
near 100 corresponds to a high stress level, and that near zero
corresponds to a low stress level.
4. Intelligent modeling and control in anesthesia
In general anesthesia, the clinical anesthesiologists’ major
challenge is to maintain the patient anesthesia via drug-induced
unconsciousness, muscle relaxation, and analgesia (i.e., pain
relief). The former two take place in the operating room,
J. Med. Biol. Eng., Vol. 32 No. 5 2012 298
whereas the latter is mainly related to postoperative conditions.
In recent years, automated drug infusion has been implemented
with feedback strategies to control the patient’s anesthesia.
However, measurement of anesthesia is the difficult part in
designing such systems. Feedback control can be used to keep
the patient in a desired anesthesia state with the advantages of
reduced induction time and minimum amount of drugs
delivered. Feedback control has been used for BP regulation,
inhalation drug concentration control, muscle relaxation control,
and ventilation control [95]. The first feedback controller used
in anesthesia was designed to control the delivery of
intravenous (IV) drug [96]. Feedback control has become
increasingly used in clinical medicine. Most investigators have
used feedback control with a classical proportional, integral,
and derivative (PID) control scheme. However, the human
body is a highly nonlinear system. The human body during
anesthesia might experience different situations, such as neural
muscular blockade, BP changes, and HRV. Therefore it is quite
difficult to use a traditional PID controller for monitoring
anesthesia. Neural network and fuzzy logic controllers are
suitable for controlling nonlinear systems [6,96,97]. Adaptive
intelligent control can be used to achieve balanced anesthesia.
The modeling and control of muscle relaxation, DOA, and
analgesia are reviewed below.
4.1 Modeling and control of muscle relaxation
4.1.1 Fuzzy logic
Anesthesiologists are familiar with administering bolus
injections of non-depolarizing muscle relaxants for controlling
the muscle relaxation at more or less regular intervals [98,99].
If the muscle relaxation is inadequate, anesthesiologist
promptly administer a bolus injection. This control method is
undesirable for eye or brain surgery because the slightest
movement could be dangerous. The patient is always given
more muscle relaxant than required, which may prolong the
effect. In order to obtain the optimal drug delivery rates, on-line
drug administration strategies should be developed [100].
Numerous computer control strategies for muscle relaxants
have been reported [101-106]. A major problem in designing
adequate feedback controllers for physiological systems is the
variation in the response to external stimuli among individuals.
Adaptive control schemes are considered suitable candidates to
solve this clinical problem [107].
A precise model of biological systems may not exist or it
may be too difficult to obtain. Knowledge-based systems
(KBSs) and intelligent computing systems are used for
modeling and control in medical diagnosis and treatment. A
KBS uses model-based reasoning (MBR), rule-based reasoning
(RBR), or case-based reasoning (CBR). Intelligent computing
methods (ICMs) include fuzzy logic (FL), genetic algorithms
(GAs), and artificial neural networks (ANNs). Pandey et al.
conducted a survey of knowledge-based and intelligent
computing systems applied to medicine, reviewing techniques
such as KBS, ANN, FL, and GAs [108]. Linkens et al.
implemented self-organizing fuzzy logic control (SOFLC) for
automated drug delivery during muscle relaxant anesthesia
[109]. The self-organization of the rule base has proven to
provide a robust controller for conditions such as model
uncertainty, noise contamination, and parameter changes [109].
For example, traditional type-1 fuzzy-based systems can handle
slight uncertainties within a short term, such as the imprecision
and slight noise associated with the inputs and outputs of fuzzy
logic controllers (FLCs). However, the effectiveness of type-
1-based agents deteriorates due to long-term uncertainties, such
as those related to heart transplants. Type-2 fuzzy sets can be
used to represent the inputs and outputs of the FLC to handle
short- and long-term uncertainties [110,111]. FLCs account for
the uncertainty in measured signals and can match the
significantly nonlinear and variable processes commonly
encountered in biomedical applications [100,112,113].
4.1.2 Neural networks
Previous studies have used the pharmacokinetic/
pharmacodynamic (PK/PD) model for determining the infusion
scheme (e.g., for mivacurium [114], atracurium [115], and
pancuronium [116]). The main drawback of this approach is the
need to find proper values for the model parameters. Lendl et al.
applied ANNs to adjust specific parameters without the
knowledge of the inner pharmacodynamic processes [117].
Good control systems can inexpensively be designed using
ANNs [118,119], which can easily be configured to
approximate the desired input/output mapping when pairs of
input and output values (samples) are available. Lendl et al.
[117] developed a novel control system which was tested
clinically on 35 patients. The system utilizes the EMG module
within the Datex AS/3 monitor for monitoring the muscle
blockade and a Graseby 3500 infusion pump for IV
administration of mivacurium. The monitor measures the ratio
of the first twitch intensity to the baseline; it decreases with
increasing neuromuscular blockade. The performance of this
feedback system compares favorably with that of closed-loop
controllers [109]. Mendonca et al. developed an automatic
system (Hipocrates) that controls the neuromuscular blockade
of anesthesia by the continuous infusion of non-depolarizing
muscle relaxants [120]. Hipocrates is based on adaptive, robust,
and classical control strategies as well as a wide range of noise
elimination techniques.
4.1.3 Hybrid and other systems
The conventional method of muscle-relaxant
administration is for the anesthesiologist to administer a bolus
dose, the size of which is determined from experience.
Supplements are given when necessary. Linkens et al. proposed
a system that automatically controls the level of muscle relaxant
in anaesthetized patients undergoing surgery [121]. A linearized
model of the patient was identified based on the
pharmacokinetic and pharmacodynamic parameters. The model
was utilized to develop a Smith predictor feedback controller.
Chuang et al. [122] developed a non-depolarizing
neuromuscular block controller which has two rule bases to
control the administration of cisatracurium. The first rule base is
developed using a fuzzy modeling algorithm (FMA), while the
second is developed based on experts’ clinical experience. The
performance of the two rule bases was tested using disturbances
such as set-point change, time delay change, noise
Intelligent Modeling and Control in Anesthesia 299
contamination, as well as sampling time changes. Results show
that the FMA-based rule-base controller had better performance.
4.2 Modeling and control of depth of anesthesia
4.2.1 Fuzzy logic
Many techniques have been utilized for monitoring DOA
based on clinical measurements such as EEG signals, BP,
plasma concentration of propofol, and AEP. Other non-
numerical clinical signs, such as lacrimation, pupil response,
and sweating, are also used for determining DOA. Linkens et al.
used a hierarchical structure based on fuzzy modeling to
monitor DOA using many clinical signs in the operating theatre
[123]. The first level uses numerical clinical signs, such as HR
and systolic arterial pressure (SAP), via a self-organizing
learning algorithm or a rule base derived from anesthesiologists’
experience to interpret primary depth of anesthesia (PDOA).
The second level uses non-numerical clinical signs, such as
lacrimation, pupil response, and sweating, which are combined
with the first level of PDOA to decide DOA. Linkens et al. also
proposed a self-organizing fuzzy model (SOFM) algorithm for
simulating anesthetic procedures during operation from the
induction to maintenance stages.
Schaublin et al. studied the fuzzy logic control of
mechanical ventilation during anesthesia [124]. This control
system automatically adjusts ventilatory frequency and tidal
volume to achieve and maintain the end-tidal carbon dioxide
fraction at a desired level (set-point). The system can achieve
and maintain a desired end-tidal carbon dioxide fraction during
routine anesthesia by controlling the mechanical ventilation
reliably and safely. Elkfafi et al. developed an intelligent signal
processing method for evoked potentials for anesthesia
monitoring and control [125]. The system was validated via 21
clinical trials and was found to reliably assess DOA during
anesthesia using propofol. Elkfafi et al. used a SOFM to
substitute an early auto-regressive technique with an exogenous
input model to maintain the drug propofol for anesthesia using
off-line analysis [126]. The results were sufficiently
encouraging to perform on-line clinical trials using fuzzy-logic-
based control and monitoring in an operating theatre.
Shieh et al. developed SOFLC and a hierarchical rule base
for DOA [127], a hierarchical system of on-line advisory for
controlling and monitoring DOA using self-organizing fuzzy
logic [128], and genetic fuzzy modeling and control of the BIS
for general intravenous anesthesia [129]. The first study proved
the suitability of the SOFLC system and hierarchical rule base
in clinical trials as an intelligent adviser for DOA management.
After extensive validation in the second study, the system was
implemented in on-line mode and was found to reduce the
recovery time. In the third study, a genetic fuzzy logic
controller (GFLC) and a genetic proportional integral
derivative controller (GPIDC) were simulated and compared
using a patient model based on the BIS value as a controlled
variable. The results indicated that both GFLC and GPIDC
control the BIS target better than manual control with similar
drug consumption.
Simanski et al. studied automatic drug delivery in
anesthesia [130]. A survey was conducted on the methods of
modeling, measurement, and general progress of closed-loop
control systems in anesthesia. The Rostock assistant system for
anesthesia control (RAN) was discussed, which provides
multiple-input/multiple-output (MIMO) control of the depth of
hypnosis. The effect of the system is similar to that of
neuromuscular blockade since the system controls arterial
hypotension. The results of 22 patients were presented.
Chou et al. developed a multivariable fuzzy logic/self-
organizing method for anesthesia control [131]. In operating
theatres, anesthesiologists usually adopt a specific regime to
administer anesthetic drugs during the different stages of the
operation. Therefore, an automatic closed-loop control system
cannot be realized using a fixed control system. A multi-stage
controller that can change from fixed to adaptive regimes is an
attractive solution. Studies have applied SOFLC to biomedical
systems for muscle relaxation [109,132,133], DOA [127,128],
and patient analgesia control [134]. Previous studies [135] have
simulated controlling anesthesia in operating theaters using a
multivariable SOFLC structure. The simulation results are
sufficiently encouraging for performing clinical trials.
Abdulla and Wen developed robust internal model control
for DOA [136]. The system includes a robust internal model
controller (RIMC) based on the BIS. Patient dose-response
models were developed to provide an adequate drug
administration regimen to avoid under- or over-dosing patients.
Simulation results showed that the RIMC outperformed the
traditional PID controller. The controller robustness was tested
by varying the patient model parameters.
4.2.2 Neural networks
No anesthesia monitoring system is considered to be
uniformly applicable to anesthesia. Robert et al. [137] surveyed
studies related to the monitoring of anesthesia using ANNs. It
was found that neural networks have been successfully utilized
to assess DOA indicators such as those based on hemodynamics,
pupillary reflex, anesthetic concentration, spontaneous EEG, and
AEPs.
Ortolani et al. [138] used ANNs to obtain a non-
proprietary index of the DOA from processed EEG data. Two
hundred patients who had undergone general abdominal
surgery were recruited for the trial. The results indicated that
the ANN was successfully trained to predict an anesthesia
depth index, ranging from 0 to 100, which correlates very well
with the BIS during total IV anesthesia with propofol.
Shieh et al. [139] used ANNs to simulate an entire surgical
operation during inhalational anesthesia. The patient model
includes four inputs (the patient’s age, weight, gender, and
anesthetic agent concentration) and four outputs (HR, SAP,
end-tidal anesthetic agents (ETAA), and EEG signals (i.e.,
BIS)). The performance of the patient model was based on the
minimum cost function and pharmacological reasoning. The
results showed that the approach can provide a more robust
model despite the considerable individual variation in
inhalational anesthesia among patients.
4.2.3 Hybrid and other systems
Allen and Smith [140] studied the neuro-fuzzy closed-loop
control of DOA. AEP measurements were utilized as a
J. Med. Biol. Eng., Vol. 32 No. 5 2012 300
feedback signal for the automatic closed-loop control of
general anesthesia using fuzzy logic and neural networks. AEP
is a signal derived from the EEG in response to auditory
stimulation, which may be useful as an index of the DOA. A
satisfactory input to a fuzzy logic infusion controller for the
administration of anesthetic drugs can be provided through a
simple back-propagation neural network for learning the AEP
signal index.
Single-input/single-output (SISO) pharmacokinetic-
pharmacodynamic compartment models are the standard
modeling paradigm used to describe the relationship between
input anesthetic agents and output patient endpoint variables.
Lin et al. [141] used hybrid multivariable models for predicting
human response to anesthesia. Model classes such as standard
linear time invariant (LTI) multivariable models, hybrid
multivariable models, nonlinear SISO PK-PD models, and
uncertain systems models have been introduced [141]. In
addition, hybrid multivariable models have been used to
describe the relations between inputs that include anesthesia,
disturbances, and surgical stimuli to a variety of patient output
variables. The focus of Lin et al.’s [141] work was a
comparison of hybrid multivariable models constructed using
subspace identification techniques to the more commonly used
SISO PK-PD models. More specifically, the hybrid models
were considered as switching system models, where the
underlying systems are MlMO linear state-space models over
which the patients’ responses switch based on their sedative
state.
Nunes et al. [142] used neural-fuzzy paradigms for
modeling and multivariable control in anesthesia. The
modeling of patient’s vital signs and the classification of DOA
were discussed. First, a hybrid patient model using Takagi-
Sugeno-Kang fuzzy models was developed. Second, a fuzzy
relational classifier was developed to classify a set of wavelet-
extracted features from the AEP into different levels of DOA. A
comprehensive patient model that predicts the effects of the
drugs propofol and remifentanil on DOA while monitoring
several vital signs was obtained. The patient’s model is the
basis for optimal multivariable control for administering these
two drugs simultaneously.
Mahfouf et al. [143] explored the dichotomy associated
with monitoring and control of DOA. The paper starts with a
rather conservative (simplistic) method to closed-loop control
of DOA via the monitoring of arterial BP using the clinical gold
standard (CGS) method followed by feedback control
techniques. The merits/limitations of each technique were
identified and a list of feasible techniques for future research in
monitoring and control of DOA was given.
Tan et al. [144] developed decision-oriented multi-
outcome modeling for anesthesia patients. The work addressed
the problem of real-time modeling for monitoring, predicting,
and diagnosing multiple outcomes of anesthesia patients. They
showed that the consideration of multiple outcomes is
necessary and beneficial for anesthesia management. It was
observed that each patient has very different dynamic responses
to drugs. Even for a given patient, responses to a given drug
change with time, patient conditions, and surgical conditions.
As a result, it is necessary to establish real-time MIMO models
for individual patients. The SISO model structure contains only
a few parameters and their values can be easily estimated in
real-time. The MIMO model is simplified significantly by
being split into multiple SISO models.
Dua et al. [145] studied modeling and multi-parametric
control for the delivery of anesthetic agents. They proposed
model predictive controllers (MPCs) and multi-parametric
model-based controllers for the delivery of anesthetic agents.
An MPC considers the patient’s state and the drug delivery rate
for solving the optimal control signal, which is performed at
regular time periods. The controller features simultaneous
control of cardiac output, mean arterial pressure (MAP), and
unconsciousness.
4.3 Modeling and control of analgesia
4.3.1 Fuzzy logic
Conventional patient-controlled analgesia (PCA) can
reduce drug consumption and improve pain control. Shieh et al.
studied a pain model and fuzzy logic patient-controlled
analgesia in extracorporeal shock-wave lithotripsy (ESWL)
[146]. A PCA based on a fuzzy logic system was developed for
controlling alfentanil administration, where the infusion rate is
adjusted using a look-up table. The look-up table is based on a
history of the drug demand. The results show that a
combination of PCA and fuzzy logic control outperforms
conventional PCA control.
Further work on pain intensity was published by Shieh et
al. [134,147], who used ESWL. A hierarchical controller whose
basic level controls analgesia and whose second level is an
adaptive self-learning level that improves the lower-level
control rules was proposed. Results show that most patients
experience a moderate level of pain when using the system
[134,147]. It was found that pain relief using the hierarchical
system is better than that using a conventional method. The
system was further enhanced by incorporating a fuzzy logic
model of the pain index that is based on the interpretation of
the self-titration of the drug delivery. Such system can be used
online to log the patient’s BP, temperature, pulse rate, and
respiration rate (RR).
Schubert et al. [148] developed a fuzzy system for the
regulation of the analgesic remifentanil during general
anesthesia based on the expert knowledge of anesthesiologists.
The anesthesiologists monitor vital parameters such as HR and
arterial BP to apply the analgesic drug. The HRV was
considered as an additional input to the fuzzy system. The final
controller structure is based on two controllers. The first
controller is used to regulate the remifentanil administration
using the HR and MAP as feedback signals. When the HR goes
higher than 90 bpm, the second controller administers a bolus
of remifenanil based on the measured HR. Both fuzzy
controllers are Mamdani inference systems.
4.3.2 Neural networks
Melzack discussed the concept of pain [149]. He found
that pain is not directly produced by inflammation, injury, or
diseases, but instead is combined and outputted from a widely
Intelligent Modeling and Control in Anesthesia 301
distributed neural network of the brain. Pain has been proposed
as the fifth vital sign to be entered into a patient’s chart along
with temperature, BP, HR, and RR. However, there is still no
objective measurement for the experience of pain. Hence, an
objective and reliable interpretation of the pain pattern is
important for pain measurement. Shieh et al. [150] proposed an
ensembled artificial neural network (EANN) with 10 inputs and
one output for the prediction of a visual analog scale (VAS) of
pain intensity. The system was tested on 323 patients receiving
intravenous PCA after Caesarean section. The prediction
performance of the EANN model was satisfactory based on the
root mean square error (RMSE) of training (1.6681) and testing
(1.9900) against the linear regression (LR) model of training
(1.9406) and testing (2.1617) for a VAS in the range of 0 to 10.
The results suggest that the EANN model can provide a useful
reference for clinical practice in acute pain service.
4.3.3 Hybrid and other systems
Jacobs et al. [151] investigated modeling and estimation for
patient-controlled analgesia of chronic pain. They derived a
mathematical model of PCA for the real-time prediction of the
continuous infusion of an analgesic drug (morphine-6β-
glucuronide). Webb et al. [152] investigated the use of a
computerized cold pressure test to measure analgesia. Yeh et al.
[153] investigated the pain relief demand index for
postoperative pain. Although they used a complicated bio-signal
processing algorithm (i.e., empirical mode decomposition) to
decompose the pain relief pattern of patients, the obtained
results were not close to those of a traditional clinical evaluation
of pain score (i.e., VAS).
5. Discussion
The main job of anesthesiologists is to maintain the
adequacy of anesthesia. Muscle relaxation, unconsciousness,
and analgesia characterize modern balanced anesthesia. The
responses of clinical signs to inadequate general anesthesia can
be divided into somatic and autonomic responses, as shown in
Table 1. Somatic responses can be further divided into sensory
motor responses (withdrawal and movement), and sensory
consciousness and nociception (awareness and pain). Although
movement (i.e., degree of muscle relaxation) can be measured,
the evaluation indices of awareness (e.g., BIS, AEP, entropy)
cannot reliably predict awareness during general anesthesia.
Several EEG-based monitors have been associated with a
decreased incidence of consciousness [40,154], reduced
administration of DOA drugs [41,42], and fast recovery from
anesthesia [155,156]. A study found that patients can become
aware even when BIS values are within the target range (i.e., 40
to 60), and thus concluded that the BIS monitoring should not
be used as part of standard practice of anesthesia [43]. Another
study has shown that the Narcotrend DOA monitor cannot
reliably predict awareness during general anesthesia via the
isolated forearm technique [157]. Hence, monitoring awareness
is still a challenging area. The assessment of pain (i.e.,
analgesia) is one of the main components of balanced general
anesthesia. However, pain is a very subjective experience
evoked by internal or external stimuli and it is very difficult to
measure via vital signs [158,159]. Currently, no commercial
monitoring method is available for evaluating analgesia during
general anesthesia.
Autonomic responses, which are more complicated, can be
classified into four types depending on the system: breathing,
hemodynamic, sudomotor, and hormonal. For the breathing
system, the breathing pattern can be monitored by RR. For the
hemodynamic system, the tachycardia, blood volume change,
and hypertension can be monitored by ECG, SpO2, and BP.
ECG and BP have been used to analyze the HRV and blood
pressure variability (BPV) in terms of sympathetic and
parasympathetic nerves using the fast Fourier transform (FFT)
to calculate the low/high frequency power ratio (LHR), which
is strongly related to the ANS. Several studies [160-164] have
used nonlinear algorithms (e.g., detrended fluctuation algorithm
(DFA) based on the α index and multi-scale entropy (MSE)
based on the complex index (CI)) to analyze RR intervals of
ECG signals in terms of fractal characteristics and the
complexity index of subjects. Although these studies focused
on cardiovascular diseases (e.g., congestive heart failure (CHF)
and arterial fibrillation (AF)), the DFA and MSE algorithms
can be applied to general anesthesia. Studies [165,166] have
calculated the cardiopulmonary coupling (CPC) via the product
of the coherence and cross-power of the ECG interval and RR
signals that are applied for the measurement of sleep quality.
Moreover, Huiku et al. [94] developed the SSI, which may be
applied to the measurement of analgesia. SSI has been
clinically validated [167,168], and will become commercially
available in the near future.
The third type, namely sudomotor innervations, may be
related to the cholinergic innervations of the ANS (i.e.,
sympathetic nerve) prominent in sweat glands, which causes
perspiration via the activation of muscarinic acetylcholine
receptors. Linkens et al. and Shieh et al. [123,127,128]
investigated the second level of the DOA via fuzzy logic to
combine the evaluation of sweat, lacrimation, and pupil
responses to anesthetic light. They combined this with PDOA,
which uses the HR and BP, to determine a confidence of the
depth of anesthesia. The final type, namely hormonal responses
(i.e., catecholamines, corticosteroids) are affected by the DOA
[169,170]. Hormonal responses can be accurately detected by
analyzing a blood sample via high-performance liquid
chromatography; however, they are not strongly related to the
DOA. In addition, hormonal measurements are expensive,
invasive, and time-consuming; therefore, the measurements
cannot be made in real-time and are not considered a major
parameter for determining adequate general anesthesia. Table 1
summarizes the use of clinical signs for adequate general
anesthesia. Multiple parameters should be considered for
deciding adequate general anesthesia.
Because it is difficult to interpret adequate general
anesthesia and the physiological signals and responses are
time-variant, the precise modeling and control of general
J. Med. Biol. Eng., Vol. 32 No. 5 2012 302
Table 2. Intelligent modeling and control of general anesthesia studies classified based on types of anesthesia parameter (i.e., muscle relaxation, depth
of anesthesia, and analgesia).
Types of anesthesia parameter
Related studies ([Reference]/Year)
Muscle relaxation
[6]/2001, [98]/1991, [99]/2002, [100]/1994, [101]/1980,
[102]/1987, [103]/1987, [104]/1988, [105]/1991, [106]/1991, [107]/1990, [108]/2009, [109]/1991, [114]/1993, [115]/1991,
[117]/1999, [120]/2004, [121]/1982, [122]/2009, [133]/1997
Depth of anesthesia
[95]/1992, [96]/1992, [97]/1997, [123]/1996, [124]/1996,
[125]/1997, [126]/1998, [127]/1999, [128]/2005, [129]/2006, [130]/2009, [131]/2010, [136]/2011, [137]/2002, [138]/2002,
[139]/2002, [140]/2001, [141]/2002, [142]/2005, [143]/2009,
[144]/2010, [145]/2010, [173]/2010
Analgesia [134]/2007, [146]/2002, [147]/2007, [148]/2008, [149]/1999, [150]/2010, [151]/1995, [152]/2001, [153]/2011
Table 3. Intelligent modeling and control of general anesthesia studies classified by type of intelligent algorithm.
Intelligent
algorithm type Related studies ([Reference]/Year)
Fuzzy logic
[6]/2001, [96]/1992, [99]/2002, [100]/1994, [107]/1990, [108]/2009, [109]/1991, [110]/2007, [111]/2010, [113]/1997,
[123]/1996, [124]/1996, [125]/1997, [126]/1998, [127]/1999,
[128]/2005, [129]/2006, [130]/2009, [131]/2010, [133]/1997, [134]/2007, [136]/2011, [146]/2002, [147]/2007, [148]/2008
Neural network [117]/1999, [118]/1993, [119]/1995, [120]/2004, [137]/2002, [138]/2002, [139]/2002, [149]/1999, [150]/2010
Hybrid and others
[97]/1997, [121]/1982, [122]/2009, [140]/2001, [141]/2002,
[142]/2005, [143]/2009, [144]/2010, [145]/2010, [151]/1995, [152]/2001, [153]/2011, [173]/2010
anesthesia is challenging. Quantitative approaches were initially
used but they have been found to be too complicated for
interpreting general anesthesia. Qualitative techniques,
particularly artificial intelligent approaches, have thus become
commonly used. Hybrid models (quantitative and qualitative
approaches) and hybrid intelligent algorithms (fuzzy logic,
neural networks, and genetic algorithms) have been applied to
general anesthesia. The intelligent modeling and control of
general anesthesia has been applied to different types of
anesthesia parameter (i.e., muscle relaxation, DOA, and
analgesia), as shown in Table 2. Most studies have focused on
how to interpret the DOA using expert systems, intelligent
systems (fuzzy logic and neural networks), or hybrid models
(genetic-fuzzy and neural-fuzzy paradigms). Muscle relaxation
and DOA can be measured directly and fed back for regulation
purposes, whereas analgesia cannot be directly measured.
Because of the difficulty of measuring pain, studies have
evaluated pain using PCA for ESWL and post-operation.
Patients in an ESWL operation are conscious, so they can feel
their pain intensity. The patients push a button when they require
pain relief. Similarly, patients start to feel pain after an operation.
It is thus convenient to use PCA for post-operation patients.
Table 3 lists research work related to anesthesia published
in the last two decades classified based on the type of intelligent
system algorithm. Most studies used fuzzy logic because it can
be applied to anesthesiologists’ decision-making. However,
complicated and high-dimensional MIMO systems are not easy
to model using fuzzy logic. The imprecision and instability of
fuzzy logic are disadvantages for general anesthesia. Hence,
ANNs have become used to solve imprecise problems if
learning data are sufficient. However, ANNs use hidden layers,
which hide a lot of information. Neuro-fuzzy modeling
[171-173] is an alternate form of artificial intelligence that
combines the transparency of fuzzy logic and the learning
ability of ANNs, which allow the understanding, validation, and
interpretation of the model produced by the rule base. Hybrid
approaches have received increasing attention because many
parameters are difficult to determine during the modeling or
control of general anesthesia. Future development should
consider neuro-fuzzy, genetic-fuzzy, genetic-neuro, and genetic-
neuro-fuzzy approaches for the modeling or control of general
anesthesia.
6. Conclusion and future developments
This paper provided a detailed review of the clinical and
engineering aspects of how to measure, interpret, model, and
control general anesthesia. Determining adequate general
anesthesia is very difficult due to the complicated mechanisms
of anesthesia, namely unconsciousness, amnesia, analgesia, and
akinesia. High-performance parallel computing and embedded
systems have been applied to biomedical engineering
applications, allowing complicated signal processing algorithms
to be implemented for general anesthesia. EMG, ECG, BP, EEG,
and SpO2 are good candidates for representing general
anesthesia. Therefore, a multistage hierarchical system for the
modeling and control of general anesthesia should be developed.
Intelligent Modeling and Control in Anesthesia 303
Acknowledgements
This research was financially supported by the National
Science Council (NSC) of Taiwan (NSC 99-2221-E-155-
046-MY3) and the Center for Dynamical Biomarkers and
Translational Medicine, which is sponsored by the NSC (NSC
100-2911-I-008-001).
References
[1] M. Neukirchen and P. Kienbaum, “Sympathetic nervous system: evaluation and importance for clinical general anesthesia,”
Anesthesiology, 109: 1113-1131, 2008.
[2] S. G. de Hert, “Volatile anesthetics and cardiac function,” Semin. Cardiothorac. Vasc. Anesth., 10: 33-42, 2006.
[3] G. Hedenstierna, “Contribution of multiple inert gas elimination technique to pulmonary medicine. 6. Ventilation-perfusion
relationships during anaesthesia,” Thorax, 50: 85-91, 1995.
[4] R. Lenhardt, “The effect of anesthesia on body temperature control,” Front. Biosci. (Schol. Ed.), 2: 1145-1154, 2010.
[5] A. S. Evers and C. M. Crowder, “Mechanisms of anesthesia and consciousness,” In: Barash, P. G., Cullen, B. F., Stoelting, R. K.,
Cahalan, M. and Stock, M. C. (Ed.), Clin. Anesth., New York:
Lippincott Williams & Wilkins, 95-114, 2009. [6] M. F. Abbod, D. G. von Keyserlingk, D. A. Linkens and M.
Mahfouf, “Survey of utilisation of fuzzy technology in medicine and healthcare,” Fuzzy Sets Syst., 120: 331-349, 2001.
[7] U. Rudolph and B. Antkowiak, “Molecular and neuronal
substrates for general anaesthetics,” Nat. Rev. Neurosci., 5: 709-720, 2004.
[8] H. C. Hemmings Jr, M. H. Akabas, P. A. Goldstein, J. R. Trudell, B. A. Orser and N. L. Harrison, “Emerging molecular
mechanisms of general anesthetic action,” Trends Pharmacol.
Sci., 26: 503-510, 2005. [9] M. T. Alkire, A. G. Hudetz and G. Tononi, “Consciousness and
anesthesia,” Science, 322: 876-880, 2008. [10] P. L. Purdon, E. T. Pierce, G. Bonmassar, J. Walsh, P. G. Harrell,
J. Kwo, D. Deschler, M. Barlow, R. C. Merhar, C. Lamus, C. M.
Mullaly, M. Sullivan, S. Maginnis, D. Skoniecki, H. A. Higgins and E. N. Brown, “Simultaneous electroencephalography and
functional magnetic resonance imaging of general anesthesia,” Ann. N. Y. Acad. Sci., 1157: 61-70, 2009.
[11] L. J. Velly, M. F. Rey, N. J. Bruder, F. A. Gouvitsos, T. Witjas, J.
M. Regis, J. C. Peragut and F. M. Gouin, “Differential dynamic of action on cortical and subcortical structures of anesthetic
agents during induction of anesthesia,” Anesthesiology, 107: 202-212, 2007.
[12] R. S. Schwartz, E. N. Brown, R. Lydic and N. D. Schiff,
“General anesthesia, sleep, and coma,” N. Engl. J. Med., 363: 2638-2650, 2010.
[13] M. T. Alkire, R. J. Haier, S. J. Barker, N. K. Shah, J. C. Wu and J. Y. Kao, “Cerebral metabolism during propofol anesthesia in
humans studied with positron emission tomography,”
Anesthesiology, 82: 393-403, 1995. [14] J. L. Feldman and C. A. Del Negro, “Looking for inspiration:
new perspectives on respiratory rhythm,” Nat. Rev. Neurosci., 7: 232-242, 2006.
[15] G. Kungys, J. B. Kim, S. L. Jinks, R. J. Atherley and J. F.
Antognini, “Propofol produces immobility via action in the ventral horn of the spinal cord by a GABAergic mechanism,”
Anesth. Analg., 108: 1531-1537, 2009. [16] M. Devor and V. Zalkind, “Reversible analgesia, atonia, and loss
of consciousness on bilateral intracerebral microinjection of
pentobarbital,” Pain, 94: 101-112, 2001. [17] N. D. Schiff, “Central thalamic contributions to arousal
regulation and neurological disorders of consciousness,” Ann. N. Y. Acad. Sci., 1129: 105-118, 2008.
[18] J. A. Campagna, K. W. Miller and S. A. Forman, “Mechanisms
of actions of inhaled anesthetics,” N. Engl. J. Med., 348: 2110-2124, 2003.
[19] N. P. Franks and W. R. Lieb, “Inhibitory synapses: anaesthetics set their sites on ion channels,” Nature, 389: 334-335, 1997.
[20] P. Hans, P. Y. Dewandre, J. F. Brichant and V. Bonhomme,
“Comparative effects of ketamine on Bispectral Index and spectral entropy of the electroencephalogram under sevoflurane
anaesthesia,” Br. J. Anaesth., 94: 336-340, 2005. [21] T. Sakai, H. Singh, W. D. Mi, T. Kudo and A. Matsuki, “The
effect of ketamine on clinical endpoints of hypnosis and EEG
variables during propofol infusion,” Acta Anaesthesiol. Scand., 43: 212-216, 1999.
[22] T. Kazama, K. Ikeda and K. Morita, “Reduction by fentanyl of the Cp50 values of propofol and hemodynamic responses to
various noxious stimuli,” Anesthesiology, 87: 213-227, 1997.
[23] A. M. Zbinden, S. Petersen-Felix and D. A. Thomson, “Anesthetic depth defined using multiple noxious stimuli during
isoflurane/oxygen anesthesia. II. Hemodynamic responses,” Anesthesiology, 80: 261-267, 1994.
[24] O. A. de Leon-Casasola and M. J. Lema, “Postoperative epidural
opioid analgesia: what are the choices?” Anesth. Analg., 83: 867-875, 1996.
[25] O. Z. Chi, W. Sommer and D. Jasaitis, “Power spectral analysis of EEG during sufentanil infusion in humans,” Can. J. Anaesth.,
38: 275-280, 1991.
[26] J. Bruhn, T. W. Bouillon and S. L. Shafer, “Electromyographic activity falsely elevates the bispectral index,” Anesthesiology, 92:
1485-1487, 2000. [27] B. Vivien, S. Di Maria, A. Ouattara, O. Langeron, P. Coriat and
B. Riou, “Overestimation of Bispectral Index in sedated
intensive care unit patients revealed by administration of muscle relaxant,” Anesthesiology, 99: 9-17, 2003.
[28] N. Liu, T. Chazot, I. Huybrechts, J. D. Law-Koune, L. Barvais and M. Fischler, “The influence of a muscle relaxant bolus on
bispectral and datex-ohmeda entropy values during propofol-
remifentanil induced loss of consciousness,” Anesth. Analg., 101: 1713-1718, 2005.
[29] M. Messner, U. Beese, J. Romstöck, M. Dinkel and K. Tschaikowsky, “The bispectral index declines during
neuromuscular block in fully awake persons,” Anesth. Analg., 97:
488-491, 2003. [30] A. R. Forbes, N. H. Cohen and E. I. Eger, “Pancuronium reduces
halothane requirement in man,” Anesth. Analg., 58: 497-499, 1979.
[31] F. C. Vasella, P. Frascarolo, D. R. Spahn and L. Magnusson,
“Antagonism of neuromuscular blockade but not muscle relaxation affects depth of anaesthesia,” Br. J. Anaesth., 94:
742-747, 2005. [32] B. Guignard, “Monitoring analgesia,” Best Pract. Res. Clin.
Anaesthesiol., 20: 161-180, 2006.
[33] D. A. Chernik, D. Gillings, H. Laine, J. Hendler, J. M. Silver, A. B. Davidson, E. M. Schwam and J. L. Siegel, “Validity and
reliability of the Observer's Assessment of Alertness/Sedation Scale: study with intravenous midazolam,” J. Clin.
Psychopharmacol., 10: 244-251, 1990.
[34] A. W. Gelb, K. Leslie, D. R. Stanski and S. L. Shafer, “Monitoring the depth of anesthesia,” In: Miller, R. D., Eriksson,
L. I., Fleisher, L. A., Wiener-Kronish, J. P. and Young, W. L. (Ed.), Miller's anesthesia, New York: Churchill Livingstone,
1229-1265, 2010.
[35] S. J. Galla, A. G. Rocco and L. D. Vandam, “Evaluation of the traditional signs and stages of anesthesia: An
electroencephalographic and clinical study,” Anesthesiology, 19: 328-338, 1958.
[36] T. P. Barnett, L. C. Johnson, P. Naitoh, N. Hicks and C. Nute,
“Bispectrum analysis of electroencephalogram signals during waking and sleeping,” Science, 172: 401-402, 1971.
[37] G. Dumermuth, P. J. Huber, B. Kleiner and T. Gasser, “Analysis of the interrelations between frequency bands of the EEG by
means of the bispectrum a preliminary study,”
Electroencephalogr. Clin. Neurophysiol., 31: 137-148, 1971. [38] J. C. Sigl and N. G. Chamoun, “An introduction to bispectral
analysis for the electroencephalogram,” J. Clin. Monit. Comput., 10: 392-404, 1994.
[39] I. J. Rampil, “A primer for EEG signal processing in
anesthesia,” Anesthesiology, 89: 980-1002, 1998.
J. Med. Biol. Eng., Vol. 32 No. 5 2012 304
[40] A. Ekman, M. L. Lindholm, C. Lennmarken and R. Sandin, “Reduction in the incidence of awareness using BIS
monitoring,” Acta Anaesthesiol. Scand., 48: 20-26, 2004.
[41] T. J. Gan, P. S. Glass, A. Windsor, F. Payne, C. Rosow, P. Sebel and P. Manberg, “Bispectral index monitoring allows faster
emergence and improved recovery from propofol, alfentanil, and nitrous oxide anesthesia,” Anesthesiology, 87: 808-815, 1997.
[42] E. Gürses, H. Sungurtekin, E. Tomatir and H. Dogan, “Assessing
propofol induction of anesthesia dose using bispectral index analysis,” Anesth. Analg., 98: 128-131, 2004.
[43] M. S. Avidan, L. Zhang, B. A. Burnside, K. J. Finkel, A. C. Searleman, J. A. Selvidge, L. Saager, M. S. Turner, S. Rao, M.
Bottros, C. Hantler, E. Jacobsohn and A. S. Evers, “Anesthesia
awareness and the bispectral index,” N. Engl. J. Med., 358: 1097-1108, 2008.
[44] A. J. Davidson, M. J. Kim and G. K. Sangolt, “Entropy and bispectral index during anaesthesia in children,” Anaesth.
Intensive Care, 32: 485-493, 2004.
[45] J. G. M. Klockars, A. Hiller, S. Ranta, P. Talja, M. J. van Gils and T. Taivainen, “Spectral entropy as a measure of hypnosis in
children,” Anesthesiology, 104: 708-717, 2006. [46] B. Vivien, X. Paqueron, P. Le Cosquer, O. Langeron, P. Coriat
and B. Riou, “Detection of brain death onset using the bispectral
index in severely comatose patients,” Intensive Care Med., 28: 419-425, 2002.
[47] N. Fàbregas, P. L. Gambús, R. Valero, E. J. Carrero, L. Salvador, E. Zavala and E. Ferrer, “Can bispectral index monitoring
predict recovery of consciousness in patients with severe brain
injury?” Anesthesiology, 101: 43-51, 2004. [48] B. A. Shapiro, “Bispectral Index: Better information for sedation
in the intensive care unit?” Crit. Care Med., 27: 1663-1664, 1999.
[49] J. W. Johansen, “Update on bispectral index monitoring,” Best
Pract. Res. Clin. Anaesthesiol., 20: 81-99, 2006. [50] S. M. Pincus, I. M. Gladstone and R. A. Ehrenkranz, “A
regularity statistic for medical data analysis,” J. Clin. Monit., 7: 335-345, 1991.
[51] J. Bruhn, L. E. Lehmann, H. Röpcke, T. W. Bouillon and A.
Hoeft, “Shannon entropy applied to the measurement of the electroencephalographic effects of desflurane,” Anesthesiology,
95: 30-35, 2001. [52] H. Schwilden, “Concepts of EEG processing: from power
spectrum to bispectrum, fractals, entropies and all that,” Best
Pract. Res. Clin. Anaesthesiol., 20: 31-48, 2006. [53] C. E. Shannon, “A mathematical theory of communication,” Bell
Sys. Tech. J., 27: 379-423, 623A-656A, 1948. [54] I. A. Rezek and S. J. Roberts, “Stochastic complexity measures
for physiological signal analysis,” IEEE Trans. Biomed. Eng., 45:
1186-1191, 1998. [55] P. F. White, J. Tang, G. F. Romero, R. H. Wender, R. Naruse, A.
Sloninsky and R. Kariger, “A comparison of state and response entropy versus bispectral index values during the perioperative
period,” Anesth. Analg., 102: 160-167, 2006.
[56] H. Viertiö‐ Oja, V. Maja, M. Särkelä, P. Talja, N. Tenkanen, H.
Tolvanen‐ Laakso, M. Paloheimo, A. Vakkuri, A. Yli‐ Hankala
and P. Meriläinen, “Description of the Entropy™ algorithm as
applied in the Datex‐ Ohmeda S/5™ Entropy Module,” Acta
Anaesthesiol. Scand., 48: 154-161, 2004.
[57] C. Thornton, M. P. Barrowcliffe, K. M. Konieczko, P. Ventham, C. J. Dore, D. E. F. Newton and J. G. Jones, “The auditory
evoked response as an indicator of awareness,” Br. J. Anaesth.,
63: 113-115, 1989. [58] D. E. F. Newton, C. Thornton, K. M. Konieczko, C. Jordan, N. R.
Webster, N. P. Luff, C. D. Frith and C. J. Dore, “Auditory evoked response and awareness: a study in volunteers at
sub-MAC concentrations of isoflurane,” Br. J. Anaesth., 69:
122-129, 1992. [59] C. Thornton, K. Konieczko, J. G. Jones, C. Jordan, C. J. Dore
and C. P. H. Heneghan, “Effect of surgical stimulation on the
auditory evoked response,” Br. J. Anaesth., 60: 372-378, 1988.
[60] D. Schwender, T. Rimkus, R. Haessler, S. Klasing, E. Pöppel
and K. Peter, “Effects of increasing doses of alfentanil, fentanyl and morphine on mid-latency auditory evoked potentials,” Br. J.
Anaesth., 71: 622-628, 1993.
[61] H. Mantzaridis and G. N. C. Kenny, “Auditory evoked potential index: a quantitative measure of changes in auditory evoked
potentials during general anaesthesia,” Anaesthesia, 52:
1030-1036, 1997. [62] T. Kurita, M. Doi, T. Katoh, H. Sano, S. Sato, H. Mantzaridis
and G. N. C. Kenny, “Auditory evoked potential index predicts the depth of sedation and movement in response to skin incision
during sevoflurane anesthesia,” Anesthesiology, 95: 364-370,
2001. [63] E. W. Jensen, M. Nygaard and S. W. Henneberg, “On-line
analysis of middle latency auditory evoked potentials (MLAEP) for monitoring depth of anaesthesia in laboratory rats,” Med.
Eng. Phys., 20: 722-728, 1998.
[64] M. M. R. F. Struys, E. W. Jensen, W. Smith, N. T. Smith, I. Rampil, F. J. E. Dumortier, C. Mestach and E. P. Mortier,
“Performance of the ARX-derived auditory evoked potential index as an indicator of anesthetic depth: A comparison with
Bispectral Index and hemodynamic measures during propofol
administration,” Anesthesiology, 96: 803-816, 2002. [65] G. N. Schmidt, P. Bischoff, T. Standl, M. Issleib, M. Voigt and J.
Schulte am Esch, “ARX-derived auditory evoked potential index and bispectral index during the induction of anesthesia with
propofol and remifentanil,” Anesth. Analg., 97: 139-144, 2003.
[66] I. A. Iselin-Chaves, H. E. El Moalem, T. J. Gan, B. Ginsberg and P. S. A. Glass, “Changes in the auditory evoked potentials and
the bispectral index following propofol or propofol and alfentanil,” Anesthesiology, 92: 1300-1310, 2000.
[67] J. Liu, H. Singh and P. F. White, “Electroencephalogram
bispectral analysis predicts the depth of midazolam‐ induced
sedation,” Anesthesiology, 84: 64-69, 1996.
[68] J. Liu, H. Singh and P. F. White, “Electroencephalographic bispectral index correlates with intraoperative recall and depth
of propofol-induced sedation,” Anesth. Analg., 84: 185-189,
1997. [69] L. A. Jr. Kearse, C. Rosow, A. Zaslavsky, P. R. N. Connors, M.
Dershwitz and W. Denman, “Bispectral analysis of the electroencephalogram predicts conscious processing of
information during propofol sedation and hypnosis,”
Anesthesiology, 88: 25-34, 1998. [70] M. M. R. F. Struys, H. Vereecke, A. Moerman, E. W. Jensen, D.
Verhaeghen, N. De Neve, F. J. E. Dumortier and E. P. Mortier, “Ability of the bispectral index, autoregressive modelling with
exogenous input-derived auditory evoked potentials, and
predicted propofol concentrations to measure patient responsiveness during anesthesia with propofol and
remifentanil,” Anesthesiology, 99: 802-812, 2003. [71] J. Bruhn, T. W. Bouillon, L. Radulescu, A. Hoeft, E. Bertaccini
and S. L. Shafer, “Correlation of approximate entropy, bispectral
index, and spectral edge frequency 95 (SEF95) with clinical signs of “anesthetic depth” during coadministration of propofol
and remifentanil,” Anesthesiology, 98: 621-627, 2003. [72] V. Bonhomme, G. Plourde, P. Meuret, P. Fiset and S. B.
Backman, “Auditory steady-state response and bispectral index
for assessing level of consciousness during propofol sedation and hypnosis,” Anesth. Analg., 91: 1398-1403, 2000.
[73] T. Katoh, A. Suzuki and K. Ikeda, “Electroencephalographic derivatives as a tool for predicting the depth of sedation and
anesthesia induced by sevoflurane,” Anesthesiology, 88:
642-650, 1998. [74] T. Katoh, H. Bito and S. Sato, “Influence of age on hypnotic
requirement, bispectral index, and 95% spectral edge frequency associated with sedation induced by sevoflurane,”
Anesthesiology, 92: 55-61, 2000.
[75] R. E. Anderson, G. Barr, H. Assareh and J. Jakobsson, “The
AAI™ index, the BIS index and end‐ tidal concentration during
wash in and wash out of sevoflurane,” Anaesthesia, 58: 531-535,
2003. [76] M. Suzuki, H. L. Edmonds, K. Tsueda, A. L. Malkani and C. S.
Roberts, “Effect of ketamine on bispectral index and levels of sedation,” J. Clin. Monit. Comput., 14: 373, 1998.
[77] J. F. A. Hendrickx, E. I. Eger, J. M. Sonner and S. L. Shafer, “Is
synergy the rule? A review of anesthetic interactions producing hypnosis and immobility,” Anesth. Analg., 107: 494-506, 2008.
Intelligent Modeling and Control in Anesthesia 305
[78] G. Barr, J. G. Jakobsson, A. Owall and R. E. Anderson, “Nitrous oxide does not alter bispectral index: study with nitrous oxide as
sole agent and as an adjunct to i.v. anaesthesia,” Br. J. Anaesth.,
82: 827-830, 1999. [79] C. Coste, B. Guignard, C. Menigaux and M. Chauvin, “Nitrous
oxide prevents movement during orotracheal intubation without affecting BIS value,” Anesth. Analg., 91: 130-135, 2000.
[80] D. Kelly and S. J. Brull, “Monitoring of neuromuscular function
in the clinical setting,” Yale J. Biol. Med., 66: 473-489, 1993. [81] S. L. Crofts and G. L. Hutchinson, “Clinical monitoring of
neuromuscular function,” Br. J. Hosp. Med., 48: 633-640, 1992. [82] J. J. Ariño-Irujo, A. Calbet-Mañueco, P. A. De la Calle-
Elguezabal, J. M. Velasco-Barrio, F. López-Timoneda, J. R.
Ortiz-Gómez, J. Fabregat-López, F. J. Palacio-Abizanda, I. Fornet-Ruiz and J. Pérez-Cajaraville, “Neuromuscular blockade
monitoring. Part 1,” Rev. Esp. Anestesiol. Reanim., 57: 153-160, 2010.
[83] J. R. Ortiz-Gómez, J. Fabregat-López, F. J. Palacio-Abizanda, I.
Fornet-Ruiz, J. Pérez-Cajaraville, J. J. Ariño-Irujo, A. Calbet- Mañueco, P. A. De la Calle-Elguezabal, J. M. Velasco-Barrio
and F. López-Timoneda, “Neuromuscular blockade monitoring. Part 2,” Rev. Esp. Anestesiol. Reanim., 57: 161-172, 2010.
[84] E. R. J. Seitsonen, I. K. J. Korhonen, M. J. Van Gils, M. Huiku, J.
M. P. Lötjönen, K. T. Korttila and A. M. Yli‐ Hankala, “EEG
spectral entropy, heart rate, photoplethysmography and motor
responses to skin incision during sevoflurane anaesthesia,” Acta Anaesthesiol. Scand., 49: 284-292, 2005.
[85] M. Rantanen, A. Yli-Hankala, M. Van Gils, H. Yppärilä-Wolters,
P. Takala, M. Huiku, M. Kymäläinen, E. Seitsonen and I. Korhonen, “Novel multiparameter approach for measurement of
nociception at skin incision during general anaesthesia,” Br. J. Anaesth., 96: 367-376, 2006.
[86] I. Takamatsu, M. Ozaki and T. Kazama, “Entropy indices vs the
bispectral index™ for estimating nociception during sevoflurane anaesthesia,” Br. J. Anaesth., 96: 620-626, 2006.
[87] H. Storm, K. Myre, M. Rostrup, O. Stokland, M. D. Lien and J. C. Raeder, “Skin conductance correlates with perioperative
stress,” Acta Anaesthesiol. Scand., 46: 887-895, 2002.
[88] M. D. Larson, A. Kurz, D. I. Sessler, M. Dechert, A. R. Bjorksten and F. Tayefeh, “Alfentanil blocks reflex pupillary
dilation in response to noxious stimulation but does not diminish the light reflex,” Anesthesiology, 87: 849-855, 1997.
[89] L. G. Kevin, A. J. Cunningham and C. Bolger, “Comparison of
ocular microtremor and bispectral index during sevoflurane anaesthesia,” Br. J. Anaesth., 89: 551-555, 2002.
[90] A. C. Morales Martín, L. M. Vaquero Roncero and C. Muriel Villoria, “Extremely prolonged neuromuscular blockade after
rocuronium: a case report,” Acta Anaesthesiol. Scand., 53:
957-959, 2009.
[91] I. Korhonen and A. Yli‐ Hankala, “Photoplethysmography and
nociception: review article,” Acta Anaesthesiol. Scand., 53:
975-985, 2009.
[92] M. Luginbühl, F. Reichlin, G. H. Sigurdsson, A. M. Zbinden and
S. Peterson-Felix, “Prediction of the haemodynamic response to tracheal intubation: comparison of laser-Doppler skin vasomotor
reflex and pulse wave reflex,” Br. J. Anaesth., 89: 389-397,
2002. [93] O. Shimoda, Y. Ikuta, M. Sakamoto and H. Terasaki, “Skin
vasomotor reflex predicts circulatory responses to laryngoscopy and intubation,” Anesthesiology, 88: 297-304, 1998.
[94] M. Huiku, K. Uutela, M. Van Gils, I. Korhonen, M. Kymäläinen,
P. Meriläinen, M. Paloheimo, M. Rantanen, P. Takala, H. Viertiö-Oja and A. Yli-Hankala, “Assessment of surgical stress
during general anaesthesia,” Br. J. Anaesth., 98: 447-455, 2007. [95] D. A. O'Hara, D. K. Bogen and A. Noordergraaf, “The use of
computers for controlling the delivery of anesthesia,”
Anesthesiology, 77: 563-581, 1992. [96] D. A. Linkens, “Adaptive and intelligent control in anesthesia,”
IEEE Control Syst., 12: 6-11, 1992. [97] D. R. Westenskow, “Fundamentals of feedback control: PID,
fuzzy logic, and neural networks,” J. Clin. Anesth., 9: 33S-35S,
1997. [98] N. Lan, P. E. Crago and H. J. Chizeck, “Feedback control
methods for task regulation by electrical stimulation of muscles,” IEEE Trans. Biomed. Eng., 38: 1213-1223, 1991.
[99] P. A. P. Fazendeiro and J. Valente de Oliveira, “A survey of fuzzy control strategies for neuromuscular blockade using
continuous infusion of atracurium,” IEEE Trans. Fuzzy Syst., 1:
547-552, 2002. [100] D. G. Mason, D. A. Linkens, M. F. Abbod, N. D. Edwards and C.
S. Reilly, “Automated delivery of muscle relaxants using fuzzy logic control,” IEEE Eng. Med. Biol. Mag., 16: 1408-1409,
1994.
[101] B. H. Brown, R. Perks, M. Anthony, J. Asbury and D. A. Linkens, “Closed-loop control of muscle relaxation during
surgery,” Clin. Phys. Physiol. Meas., 1: 203-210, 1980. [102] R. R. Jaklitsch and D. R. Westenskow, “A model-based
self-adjusting two-phase controller for vecuronium-induced
muscle relaxation during anesthesia,” IEEE Trans. Biomed. Eng., 34: 583-594, 1987.
[103] C. M. Wait, V. A. Goat and C. E. Blogg, “Feedback control of neuromuscular blockade. A simple system for infusion of
atracurium,” Anaesthesia, 42: 1212-1217, 1987.
[104] P. C. Uys, D. F. Morrell, H. S. Bradlow and L. B. Rametti, “Self-tuning, microprocessor-based closed-loop control of
atracurium-induced neuromuscular blockade,” Br. J. Anaesth., 61: 685-692, 1988.
[105] D. A. O'Hara, G. J. Derbyshire, F. J. Overdyk, D. K. Bogen and
B. E. Marshall, “Closed-loop infusion of atracurium with four different anesthetic techniques,” Anesthesiology, 74: 258-263,
1991. [106] H. Schwilden and K. T. Olkkola, “Use of a pharmacokinetic-
dynamic model for the automatic feedback control of
atracurium,” Eur. J. Clin. Pharmacol., 40: 293-296, 1991. [107] M. Denai, D. A. Linkens, A. J. Asbury, A. D. MacLeod and W.
M. Gray, “Self-tuning PID control of atracurium-induced muscle relaxation in surgical patients,” IEE Proc.-D, 137: 261-272,
1990.
[108] B. Pandey and R. B. Mishra, “Knowledge and intelligent computing system in medicine,” Comput. Biol. Med., 39:
215-230, 2009. [109] D. A. Linkens and S. B. Hasnain, “Self-organising fuzzy logic
control and application to muscle relaxant anaesthesia,” IEE
Proc.-D, 138: 274-284, 1991. [110] H. Hagras, F. Doctor, V. Callaghan and A. Lopez, “An
incremental adaptive life long learning approach for type-2 fuzzy embedded agents in ambient intelligent environments,”
IEEE Trans. Fuzzy Syst., 15: 41-55, 2007.
[111] C. Wagner and H. Hagras, “Toward general type-2 fuzzy logic systems based on zslices,” IEEE Trans. Fuzzy Syst., 18: 637-660,
2010. [112] K. Becker, B. Thull, H. Käsmacher-Leidinger, J. Stemmer, G.
Rau, G. Kalff and H. J. Zimmermann, “Design and validation of
an intelligent patient monitoring and alarm system based on a fuzzy logic process model,” Artif. Intell. Med., 11: 33-53, 1997.
[113] F. Steimann, “Fuzzy set theory in medicine,” Artif. Intell. Med., 11: 1-7, 1997.
[114] O. A. Meretoja and K. T. Olkkola, “Pharmacodynamics of
mivacurium in children, using a computer-controlled infusion,” Br. J. Anaesth., 71: 232-237, 1993.
[115] K. T. Olkkola, H. Schwilden and C. Apffelstaedt, “Model-based adaptive closed-loop feedback control of atracurium-induced
neuromuscular blockade,” Acta Anaesthesiol. Scand., 35:
420-423, 1991. [116] S. E. Kern and D. R. Westenskow, “Pharmacokinetic-based
minibolus delivery as an alternative to continuous infusion for drugs that exhibit a biophase lag,” J. Pharmacokinet. Biopharm.,
25: 191-208, 1997.
[117] M. Lendl, U. H. Schwarz, H. J. Romeiser, R. Unbehauen, M. Georgieff and G. F. Geldner, “Nonlinear model-based predictive
control of non-depolarizing muscle relaxants using neural networks,” J. Clin. Monit. Comput., 15: 271-278, 1999.
[118] A. Cochocki and R. Unbehauen, Neural networks for
optimization and signal processing, New York: John Wiley & Sons, Inc., 1993.
[119] S. S. Cross, R. F. Harrison and R. L. Kennedy, “Introduction to neural networks,” Lancet, 346: 1075-1079, 1995.
[120] T. Mendonça, H. M. Lic, P. Lago and S. Esteves, “Hipocrates: a
robust system for the control of neuromuscular blockade,” J. Clin. Monit. Comput., 18: 265-273, 2004.
J. Med. Biol. Eng., Vol. 32 No. 5 2012 306
[121] D. A. Linkens, A. J. Asbury, S. J. Rimmer and M. Menad, “Identification and control of muscle-relaxant anaesthesia,” IEE
Proc.-D, 129: 136-141, 1982.
[122] C. Chuang, S. Fan and J. Shieh, “Rule extraction by fuzzy
modeling algorithm for fuzzy logic control of cisatracurium as a
neuromuscular block,” Eng. Appl. Artif. Intell., 22: 129-140,
2009.
[123] D. A. Linkens, J. S. Shieh and J. E. Peacock, “Hierarchical fuzzy
modelling for monitoring depth of anaesthesia,” Fuzzy Sets Syst., 79: 43-57, 1996.
[124] J. Schäublin, M. Derighetti, P. Feigenwinter, S. Petersen-Felix and A. M. Zbinden, “Fuzzy logic control of mechanical
ventilation during anaesthesia,” Br. J. Anaesth., 77: 636-641,
1996. [125] M. Elkfafi, J. S. Shieh, D. A. Linkens and J. E. Peacock,
“Intelligent signal processing of evoked potentials for anaesthesia monitoring and control,” IEE Proc.- Control Theory
Appl., 144: 354-360, 1997.
[126] M. Elkfafi, J. S. Shieh, D. A. Linkens and J. E. Peacock, “Fuzzy logic for auditory evoked response monitoring and control of
depth of anaesthesia,” Fuzzy Sets Syst., 100: 29-43, 1998.
[127] J. S. Shieh, D. A. Linkens and J. E. Peacock, “Hierarchical rule-based and self-organizing fuzzy logic control for depth of
anaesthesia,” IEEE Trans. Syst. Man Cybern., 29: 98-109, 1999. [128] J. Shieh, D. Linkens and A. Asbury, “A hierarchical system of
on-line advisory for monitoring and controlling the depth of
anaesthesia using self-organizing fuzzy logic,” Eng. Appl. Artif. Intell., 18: 307-316, 2005.
[129] J. S. Shieh, M. H. Kao and C. C. Liu, “Genetic fuzzy modelling and control of bispectral index (BIS) for general intravenous
anaesthesia,” Med. Eng. Phys., 28: 134-148, 2006.
[130] O. Simanski, M. Janda, A. Schubert, J. Bajorat, R. Hofmockel and B. Lampe, “Progress of automatic drug delivery in
anaesthesia-The ‘Rostock assistant system for anaesthesia control (RAN)’,” Int. J. Adapt. Control Signal Process., 23:
504-521, 2009.
[131] Y. C. Chou, M. F. Abbod, J. S. Shieh and C. Y. Hsu, “Multivariable fuzzy logic/self-organizing for anaesthesia
control,” J. Med. Biol. Eng., 30: 297-306, 2010. [132] D. G. Mason, J. J. Ross, N. D. Edwards, D. A. Linkens and C. S.
Reilly, “Self-learning fuzzy control of atracurium-induced
neuromuscular block during surgery,” Med. Biol. Eng. Comput., 35: 498-503, 1997.
[133] J. J. Ross, D. G. Mason, D. A. Linkens and N. D. Edwards, “Self-learning fuzzy logic control of neuromuscular block,” Br. J.
Anaesth., 78: 412-415, 1997.
[134] J. S. Shieh, L. W. Chang, T. C. Yang and C. C. Liu, “An enhanced patient controlled analgesia (EPCA) for the
extracorporeal shock wave lithotripsy (ESWL),” Biomed. Eng.-Appl. Basis Commun., 19: 7-17, 2007.
[135] J. S. Shieh, M. F. Abbod, C. Y. Hsu, S. J. Huang, Y. Y. Han and S.
Z. Fan, “Monitoring and control of anesthesia using
multivariable self-organizing fuzzy logic structure,” In: Jin, Y.
and Wang, L. (Ed.), Fuzzy Systems in Bioinformatics and Computational Biology, Series: Studies in Fuzziness and Soft
Computing, Heidelberg: Springer Berlin, 273-295, 2009.
[136] S. Abdulla and P. Wen, “Robust internal model control for depth of anaesthesia,” Int. J. Mechatron. Autom., 1: 1-8, 2011.
[137] C. Robert, P. Karasinski, C. D. Arreto and J. F. Gaudy, “Monitoring anesthesia using neural networks: a survey,” J. Clin.
Monit. Comput., 17: 259-267, 2002.
[138] O. Ortolani, A. Conti, A. Di Filippo, C. Adembri, E. Moraldi, A. Evangelisti, M. Maggini and S. J. Roberts, “EEG signal
processing in anaesthesia. Use of a neural network technique for monitoring depth of anaesthesia,” Br. J. Anaesth., 88: 644-648,
2002.
[139] J. S. Shieh, S. Z. Fan and W. L. Shi, “The intelligent model of a patient using artificial neural networks for inhalational
anaesthesia,” J. Chin. Inst. Chem. Eng., 33: 609-620, 2002. [140] R. Allen and D. Smith, “Neuro-fuzzy closed-loop control of
depth of anaesthesia,” Artif. Intell. Med., 21: 185-191, 2001.
[141] H. H. Lin, C. Beck, M. Bloom and J. Chudoung, “On the use of hybrid multivariable models for predicting human response to
anesthesia,” IEEE Decision and Control, 1: 910-916, 2002.
[142] C. S. Nunes, M. Mahfouf, D. A. Linkens and J. E. Peacock, “Modelling and multivariable control in anaesthesia using
neural-fuzzy paradigms. Part I. Classification of depth of
anaesthesia and development of a patient model,” Artif. Intell. Med., 35: 195-206, 2005.
[143] M. Mahfouf, J. Backory, M. F. Abbod, C. S. Nunes and D. A. Linkens, “Monitoring and control of unconsciousness during
clinical surgery: Auditory evoked potentials (AEP) versus the
clinical gold standard (CGS),” Int. J. Adapt. Control Signal Process., 23: 522-540, 2009.
[144] Z. Tan, R. Kaddoum, L. Y. Wang and H. Wang, “Decision- oriented multi-outcome modeling for anesthesia patients,” Open
Biomed. Eng. J., 4: 113-122, 2010.
[145] P. Dua, V. Dua and E. N. Pistikopoulos, “Modelling and multi-parametric control for delivery of anaesthetic agents,”
Med. Biol. Eng. Comput., 48: 543-553, 2010. [146] J. S. Shieh, L. W. Chang, M. S. Wang, Y. P. Wang, Y. P. Yang
and W. Z. Sun, “Pain model and fuzzy logic patient-controlled
analgesia in shock-wave lithotripsy,” Med. Biol. Eng. Comput., 40: 128-136, 2002.
[147] J. S. Shieh, C. Y. Dai, Y. R. Wen and W. Z. Sun, “A novel fuzzy pain demand index derived from patient-controlled analgesia for
postoperative pain,” IEEE Trans. Biomed. Eng., 54: 2123-2132,
2007. [148] A. U. Schubert, M. Janda, O. Simanski, J. Bajorat, B. Pohl, R.
Hofmockel and B. Lampe, “A fuzzy system for regulation of the analgesic remifentanil during general anaesthesia,” IEEE
Control and Automation, 1634-1639, 2008.
[149] R. Melzack, “Pain-an overview,” Acta Anaesthesiol. Scand., 43: 880-884, 1999.
[150] J. S. Shieh, C. K. Wu, M. F. Abbod, Y. R. Wen and W. Z. Sun, “Visual analog scale of pain intensity prediction based on
ensembled artificial neural networks via i-pain system,”
Advances in Data Mining 10th Industrial Conference, 27-31, 2010.
[151] O. L. Jacobs, Y. P. Liu and H. J. McQuay, “Modeling and estimation for patient controlled analgesia of chronic pain,”
IEEE Trans. Biomed. Eng., 42: 477-485, 1995.
[152] J. A. Webb, A. Rostami-Hodjegan, R. Abdul-Manap, U. Hofmann, G. Mikus and F. Kamali, “Contribution of
dihydrocodeine and dihydromorphine to analgesia following dihydrocodeine administration in man: a PK-PD modelling
analysis,” Br. J. Clin. Pharmacol., 52: 35-43, 2001.
[153] J. R. Yeh, W. Z. Sun, Y. R. Wen and J. S. Shieh, “A novel
continuous visual analog scale model derived from pain-relief
demand index via Hilbert Huang transform for postoperative pain,” J. Med. Biol. Eng., 31: 169-176, 2011.
[154] P. S. Myles, K. Leslie, J. McNeil, A. Forbes and M. T. V. Chan,
“Bispectral index monitoring to prevent awareness during anaesthesia: the B-Aware randomised controlled trial,” Lancet,
363: 1757-1763, 2004. [155] J. W. Johansen and P. S. Sebel, “Development and clinical
application of electroencephalographic bispectrum monitoring,”
Anesthesiology, 93: 1336-1344, 2000. [156] A. Vakkuri, A. Yli-Hankala, R. Sandin, S. Mustola, S. Høymork,
S. Nyblom, P. Talja, T. Sampson, M. van Gils and H. Viertiö-Oja, “Spectral entropy monitoring is associated with reduced
propofol use and faster emergence in propofol-nitrous
oxide-alfentanil anesthesia,” Anesthesiology, 103: 274-279, 2005.
[157] I. F. Russell, “The Narcotrend ‘depth of anaesthesia’monitor cannot reliably detect consciousness during general anaesthesia:
an investigation using the isolated forearm technique,” Br. J.
Anaesth., 96: 346-352, 2006. [158] M. McCaffery and C. L. Pasero, “Pain ratings: the fifth vital
sign,” Am. J. Nurs., 97: 15-16, 1997. [159] M. K. Merboth and S. Barnason, “Managing pain: the fifth vital
sign,” Nurs. Clin. North Am., 35: 375-383, 2000.
[160] E. Rodriguez, J. C. Echeverria and J. Alvarez-Ramirez, “Detrended fluctuation analysis of heart intrabeat dynamics,”
Physica A, 384: 429-438, 2007. [161] R. G. Yeh, J. S. Shieh, G. Y. Chen and C. D. Kuo, “Detrended
fluctuation analysis of short-term heart rate variability in late
pregnant women,” Auton. Neurosci-Basic Clin., 150: 122-126, 2009.
Intelligent Modeling and Control in Anesthesia 307
[162] R. G. Yeh, G. Y. Chen, J. S. Shieh and C. D. Kuo, “Parameter investigation of detrended fluctuation analysis for short-term
human heart rate variability,” J. Med. Biol. Eng., 30: 277-282,
2010. [163] M. Costa, A. L. Goldberger and C. K. Peng, “Multiscale entropy
analysis of complex physiologic time series,” Phys. Rev. Lett., 89: 068102/1-068102/4, 2002.
[164] M. Costa, A. L. Goldberger and C. K. Peng, “Multiscale entropy
analysis of biological signals,” Phys. Rev. E, 71: 021906, 2005. [165] L. H. Ibrahim, F. J. Jacono, S. R. Patel, R. J. Thomas, E. K.
Larkin, J. E. Mietus, C. K. Peng, A. L. Goldberger and S. Redline, “Heritability of abnormalities in cardiopulmonary
coupling in sleep apnea: use of an electrocardiogram-based
technique,” Sleep, 33: 643-646, 2010. [166] D. Guo, C. K. Peng, H. L. Wu, J. E. Mietus, Y. Liu, R. S. Sun
and R. J. Thomas, “ECG-derived cardiopulmonary analysis of pediatric sleep-disordered breathing,” Sleep Med., 12: 384-389,
2011.
[167] J. Ahonen, R. Jokela, K. Uutela and M. Huiku, “Surgical stress index reflects surgical stress in gynaecological laparoscopic
day-case surgery,” Br. J. Anaesth., 98: 456-461, 2007. [168] M. M. R. F. Struys, C. Vanpeteghem, M. Huiku, K. Uutela, N. B.
K. Blyaert and E. P. Mortier, “Changes in a surgical stress index
in response to standardized pain stimuli during propofol- remifentanil infusion,” Br. J. Anaesth., 99: 359-367, 2007.
[169] T. Murakawa, Y. Satoh, T. Kudo, M. Kudo, A. Matsuki and T. Oyama, “Effects of sevoflurane anesthesia and surgery on
anterior pituitary and adrenocortical function in man,” Masui.,
36: 1058-1063, 1987. [170] T. Murakawa, Y. Satoh, T. Kudo, M. Kudo, A. Matsuki and T.
Oyama, “Effects of sevoflurane anesthesia and surgery on plasma catecholamine levels,” Masui., 38: 1456-1462, 1989.
[171] J. W. F. Catto, M. F. Abbod, D. A. Linkens and F. C. Hamdy,
“Neuro-fuzzy modeling: an accurate and interpretable method for predicting bladder cancer progression,” J. Urol., 175:
474-479, 2006. [172] J. W. F. Catto, M. F. Abbod, D. A. Linkens, S. Larré, D. J.
Rosario and F. C. Hamdy, “Neurofuzzy modeling to determine
recurrence risk following radical cystectomy for nonmetastatic urothelial carcinoma of the bladder,” Clin. Cancer Res., 15:
3150-3155, 2009. [173] H. S. Wu, H. Y. Hsu, C. C. Chang and T. C. Hsiao, “Using
neurofuzzy networks to mimic anesthesiologist knowledge in
decision making on propofol administration,” Biomed. Eng.- Appl. Basis Commun., 22: 453-464, 2010.
J. Med. Biol. Eng., Vol. 32 No. 5 2012 308