Monitoring evoked potentials during surgery to assess the level of anaesthesia

Machine learning for a combined electroencephalographic anesthesia index to detect awareness under anesthesia

Spontaneous electroencephalogram (EEG) and auditory evoked potentials (AEP) have been suggested to monitor the level of consciousness during anesthesia. As both signals reflect different neuronal pathways, a combination of parameters from both signals may provide broader information about the brain status during anesthesia. Appropriate parameter selection and combination to a single index is crucial to take advantage of this potential. The field of machine learning offers algorithms for both parameter selection and combination. In this study, several established machine learning approaches including a method for the selection of suitable signal parameters and classification algorithms are applied to construct an index which predicts responsiveness in anesthetized patients. The present analysis considers several classification algorithms, among those support vector machines, artificial neural networks and Bayesian learning algorithms. On the basis of data from the transition between consciousness and unconsciousness, a combination of EEG and AEP signal parameters developed with automated methods provides a maximum prediction probability of 0.935, which is higher than 0.916 (for EEG parameters) and 0.880 (for AEP parameters) using a cross-validation approach. This suggests that machine learning techniques can successfully be applied to develop an improved combined EEG and AEP parameter to separate consciousness from unconsciousness.

Hypnosis regulation in propofol anaesthesia employing super-twisting sliding mode control to compensate variability dynamics

Regulation of hypnosis level on bi-spectral index monitor (BIS) during a surgical procedure in propofol anaesthesia administration is a challenging task for an anaesthesiologist in multi-tasking environment of the operation theater. Automation in anaesthesia has the potential to solve issues arising from manual administration. Automation in anaesthesia is based on developing the three-compartmental model including pharmacokinetics and pharmacodynamic of the silico patients. This study focuses on regulation of the hypnosis level in the presence of surgical stimulus including skin incision, surgical diathermy and laryngoscopy as well as inter-patient variability by designing super-twisting sliding mode control (STSMC). The depth of the hypnosis level is maintained to 50 on the BIS level in the maintenance phase after improving the induction phase to 60 s using the conventional sliding mode control and 30 s with STSMC. The proposed scheme also compensates the inter-patient variability dynamics including height, age and weight of the different silico patients. Moreover, the surgical stimuli direct the hypnosis level towards the state of consciousness and stimulate the controller to provide continuous drug infusion during the interval 80-90 s. Simulation results witness that the oscillatory behaviour is observed in drug infusion to ensure the moderate level of hypnosis (40-60) for general surgery.

Differences between state entropy and bispectral index during analysis of identical electroencephalogram signals: a comparison with two randomised anaesthetic techniques

BACKGROUND

It is claimed that bispectral index (BIS) and state entropy reflect an identical clinical spectrum, the hypnotic component of anaesthesia. So far, it is not known to what extent different devices display similar index values while processing identical electroencephalogram (EEG) signals.

OBJECTIVE

To compare BIS and state entropy during analysis of identical EEG data. Inspection of raw EEG input to detect potential causes of erroneous index calculation.

DESIGN

Offline re-analysis of EEG data from a randomised, single-centre controlled trial using the Entropy Module and an Aspect A-2000 monitor.

SETTING

Klinikum rechts der Isar, Technische Universität München, Munich.

PATIENTS

Forty adult patients undergoing elective surgery under general anaesthesia.

INTERVENTIONS

Blocked randomisation of 20 patients per anaesthetic group (sevoflurane/remifentanil or propofol/remifentanil). Isolated forearm technique for differentiation between consciousness and unconsciousness.

MAIN OUTCOME MEASURES

Prediction probability (PK) of state entropy to discriminate consciousness from unconsciousness. Correlation and agreement between state entropy and BIS from deep to light hypnosis. Analysis of raw EEG compared with index values that are in conflict with clinical examination, with frequency measures (frequency bands/Spectral Edge Frequency 95) and visual inspection for physiological EEG patterns (e.g. beta or delta arousal), pathophysiological features such as high-frequency signals (electromyogram/high-frequency EEG or eye fluttering/saccades), different types of electro-oculogram or epileptiform EEG and technical artefacts.

RESULTS

PK of state entropy was 0.80 and of BIS 0.84; correlation coefficient of state entropy with BIS 0.78. Nine percent BIS and 14% state entropy values disagreed with clinical examination. Highest incidence of disagreement occurred after state transitions, in particular for state entropy after loss of consciousness during sevoflurane anaesthesia. EEG sequences which led to false 'conscious' index values often showed high-frequency signals and eye blinks. High-frequency EEG/electromyogram signals were pooled because a separation into EEG and fast electro-oculogram, for example eye fluttering or saccades, on the basis of a single EEG channel may not be very reliable. These signals led to higher Spectral Edge Frequency 95 and ratio of relative beta and gamma band power than EEG signals, indicating adequate unconscious classification. The frequency of other artefacts that were assignable, for example technical artefacts, movement artefacts, was negligible and they were excluded from analysis.

CONCLUSION

High-frequency signals and eye blinks may account for index values that falsely indicate consciousness. Compared with BIS, state entropy showed more false classifications of the clinical state at transition between consciousness and unconsciousness.

Monitoring depth of anesthesia utilizing a combination of electroencephalographic and standard measures

BACKGROUND

For decades, monitoring depth of anesthesia was mainly based on unspecific effects of anesthetics, for example, blood pressure, heart rate, or drug concentrations. Today, electroencephalogram-based monitors promise a more specific assessment of the brain function. To date, most approaches were focused on a "head-to-head" comparison of either electroencephalogram- or standard parameter-based monitoring. In the current study, a multimodal indicator based on a combination of both electro encephalographic and standard anesthesia monitoring parameters is defined for quantification of "anesthesia depth."

METHODS

Two hundred sixty-three adult patients from six European centers undergoing surgery with general anesthesia were assigned to 1 of 10 anesthetic combinations according to standards of the enrolling hospital. The anesthesia multimodal index of consciousness was developed using a data-driven approach, which maps standard monitoring and electroencephalographic parameters into an output indicator that separates different levels of anesthesia from awake to electroencephalographic burst suppression. Obtained results were compared with either a combination of standard monitoring parameters or the electroencephalogram-based bispectral index.

RESULTS

The anesthesia multimodal index of consciousness showed prediction probability (P(K)) of 0.96 (95% CI, 0.95 to 0.97) to separate different levels of anesthesia (wakefulness to burst suppression), whereas the bispectral index had significantly lower PK of 0.80 (0.76 to 0.81) at corrected threshold P value of less than 0.05. At the transition between consciousness and unconsciousness, anesthesia multimodal index of consciousness yielded a PK of 0.88 (0.85 to 0.91).

CONCLUSION

A multimodal integration of both standard monitoring and electroencephalographic parameters may more precisely reflect the level of anesthesia compared with monitoring based on one of these aspects alone.

A combination of electroencephalogram and auditory evoked potentials separates different levels of anesthesia in volunteers

BACKGROUND

It has been shown that the combination of electroencephalogram (EEG) and auditory evoked potentials (AEP) allows a good separation of consciousness from unconsciousness. In the present study, we sought a combined EEG/AEP indicator that allows both separation of consciousness from unconsciousness and discrimination among different levels of sedation and hypnosis over a wider range of anesthesia.

METHODS

Fifteen unpremedicated volunteers received mono-anesthesia with sevoflurane or propofol in a randomized crossover design in two consecutive sessions. Loss of consciousness (LOC) and EEG burst suppression (BSP) defined end-points from the upper and lower range of general anesthesia. In addition to those two extremes, the difference between anesthetic concentration at BSP and LOC was divided into three equal intervals, resulting in two intermediate levels which divided the concentration from LOC (minimum) to BSP (maximum) into three equal steps. This data set was used to test whether a previously described combined EEG/AEP indicator "detector of consciousness" can also discriminate among degrees of anesthetic effects from the awake state to BSP. Furthermore, a new improved combined EEG/AEP indicator was developed on the basis of the data from the current study, and subsequently this new indicator was tested for its ability to separate consciousness from unconsciousness with the patient data set.

RESULTS

The former "detector of consciousness" showed a prediction probability (P(K)) of 0.77 to separate different levels of anesthesia from the current study, whereas for the new combined EEG/AEP indicator, P(K) was 0.94. The new indicator was further applied to the previous study and achieved a P(K) of 0.89.

CONCLUSIONS

These results show that with the new indicator presented here, a combination of EEG and AEP parameters can be used to differentiate degrees of anesthetic effects over a wide range of hypnosis, from the conscious state to deep anesthesia (i.e., BSP).

The Narcotrend monitor and the electroencephalogram in propofol-induced sedation

BACKGROUND

The Narcotrend (NCT) is a one-channel electroencephalogram (EEG) monitor of the level of sedation. It is based on a visual EEG scoring system, which was developed by Loomis and modified by Kugler, to yield a visual expert classification (VEC) scheme for differentiation of six levels of sedation (A-F), which are subdivided into 16 substages. We designed the present study to test whether results of the automated classification of one-channel NCT input reflect those from VEC of five-channel EEG.

METHODS

Twelve healthy male volunteers received propofol using two different infusion regimens in a randomized, crossover design with concomitant NCT monitoring and VEC. Scoring results of NCT were compared with those of VEC.

RESULTS

During the infusion period, score differences of more than three substages were observed in 14 of 24 (= 58%) propofol administrations (4%-7% of total data). Often, the NCT indicated lighter sedation than VEC, which revealed more delta activity from nonfrontal leads. During recovery, NCT reported deeper sedation than VEC in 6 of 24 (= 25%) propofol administrations. Discordant trends (periods of at least five subsequent epochs with monotonic, but opposite trends for both NCT and VEC) were noted in 9 of 24 propofol administrations (37%). Furthermore, NCT had several periods when no staging information was displayed, varying from a few seconds to 10 min.

CONCLUSIONS

As the algorithm of NCT is proprietary and not accessible to the public, reasons for the observed differences between NCT and VEC cannot be analyzed and explanations must remain speculative.

NeuMonD: a tool for the development of new indicators of anaesthetic effect

Electroencephalogram (EEG) signals and auditory evoked potentials (AEPs) have been suggested as a measure of depth of anaesthesia, because they reflect activity of the main target organ of anaesthesia, the brain. The online signal processing module NeuMonD is part of a PC-based development platform for monitoring "depth" of anaesthesia using EEG and AEP data. NeuMonD allows collection of signals from different clinical monitors, and calculation and simultaneous visualisation of several potentially useful parameters indicating "depth" of anaesthesia using different signal processing methods. The main advantage of NeuMonD is the possibility of early evaluation of the performance of parameters or indicators by the anaesthetist in the clinical environment which may accelerate the process of developing new, multiparametric indicators of anaesthetic "depth".

Total intravenous anesthesia/target-controlled infusion and auditory-evoked potentials in day surgery mammoplasty

Total intravenous anesthesia and the new parameter for administering the most recent drugs, target-controlled infusion, as well as the introduction of new short half-life molecules that do not accumulate have made anesthesia in day surgery safer. In this study, the use of auditory-evoked potentials monitoring made it possible to determine the target plasma concentration of propofol that induces a narcosis sufficiently deep and strictly necessary for effectiveness, thus minimizing the anesthesiologic risk linked to the use and the dosing of the drug, reducing the hospitalization time, and decreasing the side effects for patients undergoing day surgery mammoplasty.

The influence of wavelets on multiscale analysis and parametrization of midlatency auditory evoked potentials

This work shows methodological aspects of heuristic pattern recognition in auditory evoked potentials. A linear and a nonlinear transformation based on wavelet transform are presented. They result in a statistical error model and an entropy function related to the Gibbs function and describe changes in midlatency auditory evoked potentials induced by general anaesthesia. The same transformations were calculated using 12 common wavelets. We present a method to compare the two defined parametrizations with respect to their ability to discriminate two defined states which is responsive and unresponsive depending on the wavelet used for the analysis. Auditory evoked potentials of 60 patients undergoing general anaesthesia were analysed. We propose the defined statistical error model and the entropy function as a very robust measure of changes in auditory evoked potentials. The influence of the wavelets suggest that for each parametrization the goodness of the wavelet should be validated.

Somatosensory evoked potentials as predictor of systemic inflammatory response syndrome in pigs?

OBJECTIVE

Sepsis or systemic inflammatory response syndrome (SIRS) is often associated with encephalopathy (70%), which has been described as an early symptom resulting in several diseases. The present study investigated somatosensory evoked potentials (SEP) as an indicator or even a predictor of cerebral dysfunction evaluated in an experimental model of SIRS in pigs.

METHODS

Eight Göttinger minipigs were included in the study. SIRS was mediated by induction of pancreatitis due to injection (ductus pancreaticus) of 500 mg/kg sodium taurocholate and 2.5 IU/kg enterokinase. Monitored parameters were: arterial blood-central venous-pulmonary arterial pressure, and cardiac output, systemic vascular resistance, and body temperature. SEP were recorded from centroparietal vs. frontal areas after electrical stimulation of the right forepaw.

RESULTS

At least 33% loss of vascular resistance from baseline (SIRS criteria) occurred in all animals within 4-18 h. Baseline recordings in all anesthetized animals indicated primary cortical responses to electrical stimuli identified by peak latencies between 15-20 ms (SEP(P15-20)). Attenuations in the amplitudes with significant median decreases of 46% were observed at least 4 h before the defined hemodynamic SIRS criteria.

CONCLUSIONS

The present data show a trend for the attenuation in SEP amplitudes as an indicator of systemic inflammatory response. SEP monitoring may be a sensitive marker of developing early changes in cerebral function due to SIRS-related encephalopathy.

Neuro-fuzzy closed-loop control of depth of anaesthesia

The utility of the auditory evoked potential (AEP) is under investigation as a feedback signal for the automatic closed-loop control of general anaesthesia using neural networks and fuzzy logic. The AEP is a signal derived from the electroencephalogram (EEG) in response to auditory stimulation, which may be useful as an index of the 'depth' of anaesthesia. A simple back-propagation neural network can learn the AEP and provides a satisfactory input to a fuzzy logic infusion controller for the administration of anaesthetic drugs, but the problem remains that of reliable signal acquisition.

Effects of different concentrations of sevoflurane and desflurane on subcortical somatosensory evoked responses in anaesthetized, non-stimulated patients

Twenty-four patients were recruited and given either sevoflurane or desflurane as their sole anaesthetic. Each patient was given sequentially increasing or decreasing doses at 0.5 MAC intervals, and the median nerve somatosensory evoked response recorded after an equilibration at each concentration. The N20-P25 and P25-N35 amplitudes decreased with increasing agent concentration. However, for both agents the P15-N20 amplitude response was quadratic in shape. The peak inflection points were at 3.2% for sevoflurane and 4.9% for desflurane. There were no differences between the ascending and descending groups. This increase in activity in the midbrain at 'surgical' end-tidal anaesthetic concentrations suggests more complex neuroelectrical responses to anaesthesia than simple global suppression.

Effect of tramadol on electroencephalographic and auditory-evoked response variables during light anaesthesia

Tramadol is a centrally acting opioid-like analgesic commonly used for analgesia during surgery. It has been stated that the use of tramadol increases the risk of awareness during anaesthesia. We studied 29 patients under steady state anaesthesia, ventilated via a laryngeal mask airway with 0.6 MAC isoflurane in 50% nitrous oxide, and with no surgical stimulus. The electroencephalogram (EEG) and auditory-evoked response (AER) were recorded throughout the study period, as were pulse and arterial pressure. Patients were given randomly a bolus of either saline (S), tramadol 100 mg (T1), or tramadol 200 mg (T2). Significant increases in systolic arterial pressure and decreases in heart rate were seen in the tramadol groups compared to the saline group. Significant, dose-related activation in all EEG variables (median power frequency, spectral edge, Delta Power and Alpha/Delta ratio) but no significant change in Pa or Nb amplitudes or latencies were noted. The EEG changes were not at levels thought to be associated with awareness. This study indicates that tramadol, whilst causing EEG activation, has no effect on depth of anaesthesia as measured by the AER.

Middle latency auditory evoked responses and electroencephalographic derived variables do not predict movement to noxious stimulation during 1 minimum alveolar anesthetic concentration isoflurane/nitrous oxide anesthesia

The electroencephalogram (EEG) and middle latency auditory evoked responses (MLAER) have been proposed for assessment of the depth of anesthesia. However, a reliable monitor of the adequacy of anesthesia has not yet been defined. In a multicenter study, we tested whether changes in the EEG and MLAER after a tetanic stimulus applied to the wrist could be used to predict subsequent movement in response to skin incision in patients anesthetized with 1 minimum alveolar anesthetic concentration (MAC) isoflurane in N2O. We also investigated whether the absolute values of any of these variables before skin incision was able to predict subsequent movement. After the induction of anesthesia with propofol and facilitation of tracheal intubation with succinylcholine, 82 patients received 1 MAC isoflurane (0.6%) in N2O 50% without an opioid or muscle relaxant. Spontaneous EEG and MLAER to auditory click-stimulation were recorded from a single frontoparietal electrode pair. MLAER were severely depressed at 1 MAC isoflurane. At least 20 min before skin incision, a 5-s tetanic stimulus was applied at the wrist, and the changes in EEG and MLAER were recorded. EEG and MLAER values were evaluated before and after skin incision for patients who did not move in response to tetanic stimulation. Twenty patients (24%) moved after tetanic stimulation. The changes in the EEG or MLAER variables were unable to predict which patients would move in response to skin incision. Preincision values were not different between patients who did and did not move in response to skin incision for any of the variables. MLAER amplitude increased after skin incision. We conclude that it is unlikely that linear EEG measures or MLAER variables can be of practical use in titrating isoflurane anesthesia to prevent movement in response to noxious stimulation.

IMPLICATIONS

Reliable estimation of anesthetic adequacy remains a challenge. Changes in spontaneous or auditory evoked brain activity after a brief electrical stimulus at the wrist could not be used to predict whether anesthetized patients would subsequently move at the time of surgical incision.

A system for tracking changes in the mid-latency evoked potential during anesthesia

This paper describes a method to measure changes in the mid-latency auditory evoked potential (MLAEP) during anesthesia. It is claimed that the position of the Nb-trough of the MLAEP indicates the level of consciousness. The component shows graded changes corresponding to the dose of anesthetic and it exhibits stable reproducible properties between different subjects. We propose a system that reduces the disturbances in an averaged MLAEP using fewer realizations than in the standard averaging procedure. The resulting smoothing error is reduced if the number of stimulus is decreased. Unfortunately, the variance of the waveform estimate is, thereby, increased. An improved method must be used in order to estimate the Nb-trough within a prescribed time interval of one minute. The procedure is based on inherent properties of the MLAEP and the noise. A simulation and examples of the performance on real data recorded during surgery are shown.