1. Introduction

The study of the age-dependent effects of anesthetic drugs has been a hot topic in recent years. Although general anesthesia has become a standard procedure for surgery, the age-related changes in neurophysiology and hemodynamics induced by anesthetics remain unclear.

General anesthesia induces dramatic changes in both neuronal activity and cerebral metabolism. From the perspective of neurophysiology, propofol increases frontal slow oscillation (SO) (0.1-1 Hz) and alpha rhythm (8-12 Hz) during loss of consciousness (LOC) [1,2]. The SO and delta rhythm (1-4 Hz) are also typical features of gamma-aminobutyric acid-ergic (GABAergic) general anesthesia [3,4]. While the ketamine-induced general anesthesia exhibited increased EEG power in beta (13-30 Hz) and gamma (30-48 Hz), it did not show an increase in alpha rhythm [5]. From the perspective of hemodynamics, anesthetic drugs have a variety of effects. For example, sevoflurane suppresses cerebral metabolism and has less effect on cerebral blood flow (CBF) according to previous studies in adults [6,7]. However, the effects of sevoflurane on hemodynamics in children are rarely reported. Propofol-induced general anesthesia decreases regional CBF and vasoconstriction, but preserves the metabolism’s oxygen demand/supply balances and cerebral auto-regulation [7]. Each of these variable influences of anesthetic agents suggests the need to pursue an effective tool to ensure intraoperative safety.

The electroencephalogram (EEG), as a tool to measure brain activity at a macro-scale, has become the de facto standard for monitoring the depth of anesthesia [8]. Also, the rapid brain development from birth to adolescence is marked by fundamental changes in brain connectivity which can be reflected by EEG oscillations [9–11]. This suggests that EEG features are an effective and age-adjusted measure to investigate changes in neuronal activity during general anesthesia. Beyond their neuronal effects on the brain, anesthetics have broad actions that include hemodynamic (e.g., vascular reactivity) and baseline physiological changes (e.g., heart rate and blood pressure). Functional near-infrared spectroscopy (fNIRS), as a portable, non-invasive, and cost-effective neuroimaging tool, has been developed and applied to investigate the cortical hemodynamic activity changes associated with motion, attention and memory [12,13]. Several studies have employed the raw fNIRS-derived hemodynamic variables, i.e., oxyhemoglobin (Δ[HbO]), deoxy-hemoglobin (Δ[Hb]) that are calculated using the modified Beer-Lambert Law to qualify anesthesia’s effects [6,7]. The existing studies have combined the two methods of EEG and fNIRS to analyze anesthetic effects during surgery [14–16], however little work has been done thus far to employ the synchronized multimodal EEG-fNIRS dynamics to study neurovascular coupling with age and depth of anesthesia (DoA) induced by general anesthesia.

In this study, we employed propofol-induced and sevoflurane-maintained general anesthesia. The protocol is a common anesthesia method for both children and adults in the hospital where we collected data. We explored the effects of propofol induction and sevoflurane maintenance general anesthesia in neuronal activity and hemodynamics in both children and adults. We hypothesized that there is an age-dependency between neuronal activities and hemodynamic responses in children and adults during general anesthesia, and differences among different anesthetic states. To verify our hypotheses, we analyzed the age dependence and state discrimination properties of EEG and fNIRS features. We simultaneously obtained the EEG and fNIRS data in both children and adults during general anesthesia. Then, we analyzed the age-related EEG spectrum, permutation entropy (PE), hemodynamic parameters (Δ[Hb], Δ[HbO], and Δ[HbT]), as well as the coupling between neuronal and hemodynamic parameters for children and adults in various states, including correlation, coherence and Granger-causality (GC).

2. Materials and methods

2.1 Participants and data acquisition

We recorded EEG and fNIRS signals from patients that had undergone surgical treatment at the Seventh Medical Center at the Chinese PLA General Hospital, Beijing, China. Patients were Physical Status I or II (American Society of Anesthesiologists), and clinically stable on the day of study. Patients with central neurological disorders, including autism, attention-deficit/hyperactivity disorder, epilepsy, or other congenital psychiatric conditions, were excluded from the analysis. Also, any patients who had undergone cardiac surgery were excluded from the study. Based on these criteria, children between the age range of 6-12 years (n = 17, 5 females) and adults within the age range of 18-60 years (n = 25, 10 females) were enrolled for analysis. Written informed consent was obtained from the patients themselves (adults) or their parent(s) (children), following the recommendations of the Seventh Medical Center ethics committee at the Chinese PLA General Hospital.

The procedure is shown in Figure 1(A). The data recorded include the states of wakefulness (before drug delivery and closing of eyes), induction (the period from drug delivery to LOC), maintenance of a surgical state of anesthesia (MOSSA) (the period from LOC to drug withdrawal), emergence process [the period from drug withdraw to period after recovery of consciousness (ROC)], and recovery (the period after ROC) among various age groups.

 

Fig. 1. Schematic diagram of the entire procedure (A) The procedure during general anesthesia. (B) The position of the BIS probe and the fNIRS probes. The probes on the left were selected for analysis (red dotted box). The picture was generated by FaceGen Modeller. (C) Schematic diagram of a fNIRS probe with one source and two detectors. LOC: loss of consciousness; ROC: recovery of consciousness.

Download Full Size | PDF

The EEG data was collected by the BIS monitor (Medtronic, USA). The hardware bandpass filtering range of the BIS system is 0.25-100 Hz, therefore, the SO ranges from 0.25 to 1 Hz is included in the study. The BIS monitor recorded two frontal EEGs with electrode positions near FPz-F9 and FPz-AF7 (international 10-20 system). We checked electrode impedance throughout the EEG recording and guaranteed less than 5 kΩ for each channel. The sampling rate was 128 Hz. The hemodynamic responses of Δ[HbO], Δ[Hb], and Δ[HbT] were recorded with a sampling rate of 0.5 Hz. Figures 1(B) and 1(C) show the layouts and positions of the EEG and fNIRS probes during recording. Because the EEG recordings from two channels were located in the same hemisphere of the brain affected by noise from head movement and electrotome interference almost simultaneously, and they were highly correlated (correlation coefficient = 0.76 ± 0.22), we only analyzed the EEG from FPz-F9. To study the same side of the EEG, we chose the fNIRS probe on the left for analysis.

The fNIRS system used in this study is the EGOS-600 (EnginMed, Co., Ltd. Suzhou, China), which uses 700 nm and 830 nm dual-wavelength LEDs as light emitting source, and the hemodynamic responses of Δ[HbO], Δ[Hb], and Δ[HbT] were recorded with a sampling rate of 0.5 Hz. The differential path-length factor (DPF) was set at 4.2 and 3.9 for 700 nm and 830 nm, respectively [17]. There are two sources and one detector, the distances between the two sources and the detector are 4 cm and 3 cm, respectively. The system uses the modified Beer-Lambert Law to calculate Δ[Hb] and Δ[HbO], and calculates regional cerebral oxygen saturations (rSO2) based on the spatially resolved spectroscopy (SRS) algorithm [18]. However, the system only outputs the Δ[Hb] and Δ[HbO] at the source-detector distance of 4 cm. At this distance, more oxygen metabolism in the cerebral cortex can be detected. Because there was no information of Δ[Hb] and Δ[HbO] in 3 cm or less, we cannot remove the influence of skin vasculature by using the short distance regression [19].

The criteria of application using children's electrode (the distances of two sources and detector are 3 cm and 2 cm) is the patient’s age less than 4 years old and their weight less than 18 kg. Since our children are over 6 years old and weigh more than 18 kg, we use the same type of probe for children as that in adults. Because we can only obtain the Δ[Hb] and Δ[HbO] indices from the devices but not the light intensity values, we cannot use an age-adaptive DPF to calculate the Δ[Hb] and Δ[HbO] indices for children again. So, there were no correction in calculating the Δ[HbO] and Δ[Hb] in children.

To ensure safety during surgery, the electrocardiogram, heart rate, and pulse oximetry were monitored by using an iPM9800 patient monitor (Mindray Co., Ltd. Shenzhen, China). Data were collected between December 15, 2018 and September 30, 2019. Details on the induction and maintenance drugs are provided in Table 1. Details on the type of surgery and anesthetic drugs used for each subject are provided in Table S1 (in Supplement 1).

Table 1. Clinical characteristics of all subjects studied.^a

View Table

2.2 EEG and fNIRS data preprocessing

Because it is meaningful to investigate the neurovascular coupling between neuronal activity and hemodynamics in the same brain region, we only selected the EEG and fNIRS probes on the same side for analysis.

We preprocessed the original EEG signals for offline analysis using MATLAB (version 2014, MathWorks Inc.). Most of the noise that needed to be eliminated was from head movement, power lines, baseline drift, or other physiological signals (e.g., electrooculogram (EOG) and electromyogram (EMG)). Therefore, we conducted the following signal processing pipelines: (1) diminishing 50 Hz power line noise with an adaptive notch filter [20]; (2) eliminating high frequency noise (>45 Hz) and low frequency baseline drift (<0.1 Hz) with a band-pass finite-impulse response (FIR) filter [21]; (3) detecting and eliminating EMG influence with an inverse filter; (4) removing any EOG signals, (which were only present during wakefulness and recovery state), with a wavelet-based noise removal method [22]; (5) removing any EEG amplitudes that exceeded 200 µV or were more than 3 standard deviations away from the mean (in a 10-s segment of data) and thus were considered outliers.

The EGOS-600A outputs hemodynamic indices with a 0.5 Hz sampling rate. It has been documented that hemodynamic responses below 0.1 Hz are highly related with the brain’s functional activities [23,24]. Under a 0.5 Hz sampling rate, we used kurtosis-based wavelet filtering to filter out any motion noise [25]. We used a zero-phase digital filter (Matlab function of filtfilt.m) with a Butterworth order equal to 3 for low-pass filtering.

2.3 EEG features extraction

2.3.1 Bispectral index (BIS)

BIS is a processed index of frontal EEG parameters [26]. It uses a proprietary algorithm containing fixed, weighted ratios of several subparameters based on EEG through a time-spectrum and bispectral analysis of the original signal (including beta-ratio, SynchFastSlow, QUAZI, and burst suppression rate) to generate its output through a time-spectrum and bispectral analysis of the original signal [27]. It produces a dimensionless number between 0 and 99, representing no significant drug effect and maximum hypnotic drug effect, respectively.

2.3.2 Spectral analysis

To investigate the EEG spectral characteristics in the three states and two age groups, we employed the multi-taper spectral method to calculate age-group-averaged spectrograms and EEG powers in the 0.25-45 Hz frequency band with the Chronux toolbox (version 2.11) [28,29]. The parameters of spectrograms used in the calculation were window length = 2 s with no overlap; time-bandwidth product = 2; number of tapers = 3; and spectral resolution = 0.5 Hz. We calculated the EEG powers in the frequency bands of SO (0.25-1 Hz), delta (1-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), beta (13-25 Hz), and gamma (25-45 Hz) in the three states [30]. It was suggested that relative power could distinguish different anesthesia states. Also, the alpha/delta power ratio and (alpha + beta)/(delta + theta) power ratio were utilized to assess alterations in the frequency distribution of EEG power, capturing the general shift in power from higher to lower frequencies [31,32]. In the present study, the alpha/delta power ratio and (alpha + beta)/(delta + theta) power ratio were calculated.

In order to analyze the neurovascular coupling between neural activities and hemodynamics, we analyzed the EEG power in the frequency bands of the SO, delta, theta, alpha, beta, and gamma, with an epoch of 10 s and an overlap of 80%. The power time series in these frequency bands were denoted as ${P_{SO}}$, ${P_{delta}}$, ${P_{theta}}$, ${P_{alpha}}$, ${P_{beta}}$ and ${P_{gamma}}$. Under the procedure, we can get the power time series with a sampling rate of 0.5 Hz.

2.3.3 Permutation entropy (PE) analysis

PE is a simple and robust algorithm originally proposed by Bandt and Pompe [33]. Various studies have proved that PE is a promising tool for quantifying the DoA [34–36]. The PE transforms the given time series into a series of symbolic ordinal patterns, each describing the order relations between the present and a fixed number of equidistant past values at a given time. Because neuronal activities are complex, PE has been widely used to study the EEG’s nonlinear dynamic characteristics [37,38]. Especially, some studies have shown that the PE can quantify the consciousness and monitor the DoA [35,37]. The detailed description of PE is presented in Supplement 1. We chose an embedding dimension m = 6 and a time delay τ=1 for PE calculation [35] using a fixed window length of 10 s and a 80% overlapping. Based on this parameter setting, the sampling rate of PE is 0.5 Hz, the same as the fNIRS signal.

2.4 Neurovascular coupling analysis

We analyzed the neurovascular coupling between EEG activities and hemodynamic responses from the perspectives of time domain, frequency domain, and causality.

2.4.1 Correlation analysis

In time domain, we calculated the correlation between neuronal activities (${P_{SO}}$, ${P_{delta}}$, ${P_{theta}}$, ${P_{alpha}}$, ${P_{beta}}$, ${P_{gamma}}$, and PE) and hemodynamic responses (Δ[HbO] and Δ[Hb]) in three states. We determined the correlation strength according to the correlation coefficient (r) based on Pearson's correlation via the MATLAB corrcoef function, which ranged from -1 to 1. Also, the linear regression was calculated based on the MATLAB regress function.

2.4.2 Coherence analysis

Coherence is a method to measure the linear correlation of two time series in the frequency domain. Its calculation involves using the normalized cross spectral density function to quantify the coherence between the two signals by calculating the autocorrelation power spectral density and cross power spectral density of two time series [39].

In this study, we calculated the coherence value using the coherencyc function in the Chronux toolbox. The value of coherence ranges from 0 to 1, where 0 means that two time series have no correlation or synchronization, while coherence of 1 means that two time series have exact correlation or synchronization.

In the present study, the coherence in the frequency domain ($CO(f )$) was calculated between neuronal indices (${P_{SO}}$, ${P_{theta}}$, ${P_{delta}}$, ${P_{alpha}}$, ${P_{beta}}$, ${P_{gamma}}$ and PE) and hemodynamic responses (Δ[HbO] and Δ[Hb]) with a sampling rate of 0.5 Hz. We focused on the results of coherence in the frequencies less than 0.1 Hz. The coherence indices between neuronal activities and hemodynamic responses were termed as $C{O_{x - y}}$, where x represents the neuronal indices and y represents the hemodynamic responses.

2.4.3 Granger-causality analysis

GC describes the directedness of information flow by using autoregressive models to assess whether past information helps to predict current information [40]. GC implements a statistical interpretation of causality: Y ‘Granger causes’ X if the past of Y is more helpful in predicting the future of X than the past of X alone. GC has been an effective metric to measure the effective connectivity based on EEG study [41], fNIRS [42,43] and functional magnetic resonance imaging (fMRI) [44]. Especially, GC also was employed in investigating the neurovascular coupling [45]. In our study, we employed GC to explore the neurovascular coupling between neuronal and hemodynamic activities. We conducted GC analysis with a free MATLAB toolbox-GCCA (version 2.9) [46].

Because it is a reasonable hypothesis that the neuronal activities induced the change of hemodynamics based on the neurovascular coupling theory, we only analyze the GC indices derived from EEG indices (${P_{SO}}$, ${P_{delta}}$, ${P_{theta}}$, ${P_{alpha}}$, ${P_{beta}}$, ${P_{gamma}}$, and PE) to hemodynamic responses (Δ[Hb] and Δ[HbO]) for causality investigation. We term the causality from EEG indices to hemodynamic indices as $G{C_{x \to y}}$, where x represents the EEG indices and y represents the hemodynamic indices. The detailed contents about the parameters’ selection are presented in Supplement 1.

2.5 Statistical analysis

In our study, no “a priori” statistical power calculation was conducted before the study and the sample size was based on the available number of patients’ fNIRS and EEG recordings.

R Project (version 4.2.2; http://www.r-project.org) was used for statistical analysis. The study conducted statistical analysis for EEG features, hemodynamic responses and neurovascular couplings between different age groups and states. Z-score normalization was used to normalize the values of GC. Next, a linear mixed model with Satterthwaite’s approximation was employed to analyze the interaction effects between two factors: (1) age group (two levels: children and adults), and (2) state (three levels: wakefulness, MOSSA and recovery). After obtaining the interaction effects of the two factors, the 95% confidence interval values of the estimated differences were calculated. The Tukey test is simple to use and suitable for small sample data. Finally, we used a Tukey test for post hoc testing for significant tests of age group and state. In all figures and tables, p < 0.05, p < 0.01, and p < 0.001 are denoted with ‘*’, ‘**’, and ‘***’, respectively.

3. Results

Propofol induction and sevoflurane maintenance were used in this study for all patients. We analyzed the EEG and hemodynamic responses of two age groups in the states of wakefulness, MOSSA and recovery. We chose the EEG and hemodynamics epochs that we analyzed in MOSSA after onset of surgery with no excess noise and in a stable state, as viewed in a spectrogram and checked by an anesthesiologist. The propofol bolus was administered by an anesthesiologist when deeper DoA was needed during the surgical incision. Therefore, we excluded any EEG and fNIRS recordings within 10 minutes after injection for the state analysis. We used the manual record to confirm that there was no synergy between the propofol and the sevoflurane. Additionally, we selected the recordings of the expired sevoflurane concentrations that ranged in 2.51 ± 0.41, 2.00 ± 0.26 for children and adults, respectively, during MOSSA. The minimum alveolar concentration (MAC) ranges for these two age groups during the EEG epochs that we analyzed had no significant difference (children: 1.13 ± 0.21, adults: 1.08 ± 0.12) based on the age-related MAC analysis [47] (Wilcoxon test by rank, p = 0.87). The individual investigations of a child (7-year-old) and an adult (44-year-old) patients with changes in EEG and hemodynamic response during general anesthesia were presented in Figure 2.

 

Fig. 2. The EEG, hemodynamic and neurovascular coupling changes in (A)-(H): a 7-year-old child and (I)-(P): a 44-year-old adult versus time during general anesthesia. (A) The processed electroencephalogram recording is from the FPz-F9. The sampling rate is 128 Hz. (B) The spectrogram was computed via short-time Fourier transform. The dark red indicates higher power, and the blue means lower power. (C) Time course for BIS and normalized PE indices. (D) The Δ[Hb] spectrogram was computed via short-time Fourier transform. The dark red indicates higher power, and the blue means lower power. (E) The Δ[HbO] spectrogram was computed via short-time Fourier transform. The dark red indicates higher power, and the blue means lower power. (F) The changes in Δ[Hb], Δ[HbO] and Δ[HbT] in the 0.01-0.1 Hz frequency band during general anesthesia. (G) The coherence spectrum of $C{O_{SO - Hb}}$ during general anesthesia. The black curve is the mean of the coherence spectrum. (H) The GC from permutation entropy to Δ[Hb] (blue) and Δ[HbO] (red) during general anesthesia. The $G{C_{PE \to Hb}}$ represents the Granger-causality form the permutation entropy to Δ[Hb], and the $G{C_{PE \to HbO}}$ represents the Granger-causality from the permutation entropy to Δ[HbO]. (I)-(P) are the same as (A)-(H). LOC, loss of consciousness; ROC, recovery of consciousness; EEG, electroencephalogram; SO, slow oscillation (0.25-1 Hz); BIS, bispectral index; PE, Permutation entropy; GC, Granger-causality; CO, coherence.

Download Full Size | PDF

We analyzed the EEG derived features: spectrograms, BIS, PE; fNIRS derived features: Δ[Hb] and Δ[HbO] spectrograms less than 0.25 Hz, and Δ[Hb], Δ[HbO] and Δ[HbT] in 0.01-0.1 Hz; and the neurovascular coupling indices between EEG indices and hemodynamic responses: coherence and GC. Figure 2(A)-(H) and Figure 2(I)-(P) show the preprocessed EEG recordings, spectrogram, PE and BIS, the Δ[Hb] and Δ[HbO] spectrograms, hemodynamic responses, the coherence spectrum of $C{O_{SO - Hb}}$ and the GC from PE to Δ[Hb] and Δ[HbO] ($G{C_{PE \to Hb}}$ and $G{C_{PE \to HbO}}$) for a child and an adult during general anesthesia, respectively.

The EEG power increased (i.e., the dark red area increased) in low frequencies (<15 Hz). Also, there was a distinct bifurcation of SO-delta (0.25-4 Hz) and alpha (8-13 Hz) oscillations during MOSSA for the child and the adult. The normalized PE curve was similar to the trend of the BIS during general anesthesia for the two patients. Figure 2(D), (E), (L) and (M) show the Δ[Hb] and Δ[HbO] spectrograms less than 0.25 Hz for the two patients. The frequency of the mechanically aided breathing is about 0.2 Hz [48]. Due to the existence of the effect of aliasing, it is usually necessary to set 5 times the frequency of the concerned signal to obtain effective information [49]. However, the sampling rate of the fNIRS signal is 0.5 Hz. Therefore, we could not observe the 0.2 Hz information in the Δ[Hb] and Δ[HbO] spectrograms. In our fNIRS data analysis, a bandpass filter of 0.01-0.1 Hz was used to extract the Δ[HbO] and Δ[Hb] signals. So, the respiration did not affect our fNIRS data analysis. In Figure 2(F) and (N), the Δ[HbO] and Δ[HbT] increased, while the Δ[Hb] decreased after LOC. These changes in curves are consistent with those found by Curtin et al. [7]. Meanwhile, after ROC, both the Δ[HbO] and Δ[Hb] increased for the child and decreased for the adult.

For neurovascular coupling indices, the $C{O_{SO - Hb}}$ increased during MOSSA in the frequency band of 0.01-0.1 Hz for two patients. This indicates that there was stronger synchronization between SO and Δ[Hb] after LOC. Both $G{C_{PE \to Hb}}$ and $G{C_{PE \to HbO}}$ decreased after LOC for two patients. This means that the directional connectivity from PE to hemodynamic responses decreased after LOC. However, the $G{C_{PE \to Hb}}$ and $G{C_{PE \to HbO}}$ curves have crosstalk during MOSSA. Both $G{C_{PE \to Hb}}$ and $G{C_{PE \to HbO}}$ increased after ROC, and the $G{C_{PE \to Hb}}$ indices were higher than those of $G{C_{PE \to HbO}}$ in the recovery state.

3.1 Age-related EEG and hemodynamic activities across age groups from wakefulness to recovery

For each patient, we extracted 100-s segments of EEG data recorded in 0.25-45 Hz frequency bands in each state. Figures S1(A)-(C) present the spectra for the two age groups in three states (in Supplement 1). The EEG power of children is stronger than that of adults except for the high frequency band (25-45 Hz) in the recovery (see Figure S1 (C)).

To further investigate the changes in EEG power between the two age groups, we calculated the mean and standard deviation (SD) of the absolute power at six frequency bands. Figure 3 shows the mean ± SD of the EEG power for the frequency bands across the age groups during various states. Detailed values are presented in Table S2 (in Supplement 1). It was shown that the children’s group has higher power than the adult group in most frequency bands and states except for beta and gamma during recovery.

 

Fig. 3. The EEG powers of children (yellow) and adults (pink) age groups during wakefulness, MOSSA, and recovery in six frequency bands: (A) SO, (B) delta, (C) theta, (D) alpha, (E) beta, and (F) gamma. The EEG power ratio: (G) alpha/delta and (H) (alpha + beta)/(delta + theta). The bar and error bars indicate the mean ± standard deviation. ***Significant difference in the percentage of nonzero EEG power values at P < 0.001. SO, slow oscillation (0.25-1 Hz); δ, delta rhythm (1-4 Hz); θ, theta rhythm (4-8 Hz); α, alpha rhythm (8-12 Hz); β, beta rhythm (12-30 Hz); γ, gamma rhythm (30-45 Hz).

Download Full Size | PDF

As shown in Table S3 (in Supplement 1), there were significant decreases from children to adults for most EEG powers and EEG powers ratios in the same state (p < 0.01), especially in MOSSA (EEG power in SO: < 0.05, others: < 0.001). This indicates that the EEG powers are age-dependent, especially in MOSSA. The statistical significances among the three stages in different age groups are presented in Table S4 (in Supplement 1). There are significant differences for most indices between wakefulness and MOSSA, as well as between MOSSA and recovery in the two age groups (p < 0.05), especially for EEG power in SO (p < 0.001). In the medium frequency bands (from delta to beta), most EEG powers of MOSSA were significantly higher than those of wakefulness or recovery (p < 0.001) for both age groups. While, EEG power in SO was significantly decreased from wakefulness to recovery for children and adults (p < 0.001). The alpha/delta values were significantly higher than those of wakefulness and recovery (p < 0.001) for both age groups. This means that EEG power at different frequencies can distinguish anesthesia states in both children and adults, but there are different patterns of change. Also, the alpha/delta is a good index to distinguish the anesthesia state.

In addition, we analyzed the BIS and PE derived from EEG, and the Δ[Hb], Δ[HbO] and Δ[HbT] derived from fNIRS, in two age groups. Figures 4(A)-(E) present box plots for these indices in the two age groups in the three states. For the EEG parameters, both the BIS and PE decreased after LOC and increased after ROC. For the hemodynamic parameters, only Δ[Hb] decreased in MOSSA and then increased in recovery. The mean ± SD values of the relevant indicators across age groups during various states are presented in Table S5 (in Supplement 1).

 

Fig. 4. Box plots of five indices of two age groups in wakefulness, maintenance of a surgical state of anesthesia, and recovery: (A) bispectral index, (B) permutation entropy, (C) Δ[Hb], (D) Δ[HbO] and (E) Δ[HbT]. Changes in Δ[Hb] (blue) and Δ[HbO] (red) at four-time points in the 4 minutes after induction and the four minutes before the recovery in (E) children and (F) adults. **Significant difference in the percentage of the indices at P < 0.001. BIS, bispectral index; MOSSA, maintenance of a surgical state of anesthesia; PE, Permutation entropy. ROC, recovery of consciousness. Induction, the period from drug delivery to LOC; Emergence process, the period from drug withdraw to ROC.

Download Full Size | PDF

We conducted further statistical tests to analyze the significance of these indices both between age groups in the same state, and among various states in the same age group (see Table S6 and S7 in Supplement 1). For all the EEG and hemodynamic indices, there were no significant differences between age groups in three states (p > 0.05). This indicates that the EEG and hemodynamic indices are not age-independent. While the values of EEG indices (i.e., BIS and PE) and Δ[Hb] significantly decreased from wakefulness to MOSSA, and increased from MOSSA to recovery in the two age groups (p < 0.001). This means that these indices can distinguish between MOSSA and two other states. Besides, the values of Δ[HbO] and Δ[HbT] significantly decreased from MOSSA to recovery in adults (p < 0.05). These results suggest that EEG indices and Δ[Hb] may be more promising biomarkers than Δ[HbO] and Δ[HbT] in distinguishing the anesthesia states. The changes in EEG and hemodynamic indices of adults are more significant than those of children. However, these indices are not age-dependent.

In order to investigate the hemodynamic changes during induction and emergence process in a short time span, we analyzed the changes in Δ[Hb] and Δ[HbO] at four time points during the 4 minutes after induction and before recovery in each age group (see Figure 4(F) and (G)). For children, as shown in Figure 4(F), Δ[Hb] gradually decreased, and Δ[HbO] increased, during the first 4 minutes after propofol induction. Both Δ[HbO] and Δ[Hb] are negative and trend toward the baseline during the emergence process. The adults trend similarly with the children during induction and recovery (see Figure 4(G)). However, the absolute values of Δ[Hb] in adults are less than those in children. See Table S8 (in Supplement 1) for all means, minimums, and maximums of Δ[HbO] and the Δ[Hb] indices.

3.2 Age-related neurovascular coupling in various anesthesia states

We systematically analyzed the relationship between EEG indices and hemodynamic responses by using correlation, coherence and GC measures.

Firstly, for the correlation measure, we found that absolute r values were generally not high between EEG and hemodynamic parameters. The scatter plots of EEG indices vs. Δ[Hb] (Figure 5) and Δ[HbO] (Figures S2 in Supplement 1) in the two age groups are presented. The values (black points) were taken from all subjects in each group during the time course of general anesthesia (from wakefulness to recovery) with a step of 30 s. As we can see, only the ${P_{gamma}}$ and Δ[Hb] in adults, as well as the PE and Δ[Hb] in the two age groups are positively median correlated (r > 0.3). There was a weak correlation between most of the indices (r < 0.3). It means that a weak linear correlation was between neuronal activities and hemodynamic responses, except for PE and Δ[Hb].

 

Fig. 5. Scatter plots (black points) and linear regression (red line) between EEG indices and Δ[Hb] in the two age groups. r, correlation coefficient.

Download Full Size | PDF

Secondly, we analyzed the age-dependency of neurovascular coupling in frequency domain. The heatmaps of the coherence indices between EEG indices and hemodynamic activities are shown in Figure 6(A) and (B). The asterisk indicates the significant differences between age groups. The mean ± SD for coherences in different age groups and states is shown in Table S9 and S10 (in Supplement 1). The significance tests for all of these indices between different age groups in each state are shown in Table S11 and among different states in each age group are shown in Table S12 and S13 (in Supplement 1).

 

Fig. 6. The heatmaps of coherence: (A) between neuronal activities and △[Hb], and (B) between neuronal activities and △[HbO], GC: (C) from neuronal activities to △[Hb], and (D) from neuronal activities to △[HbO]. ***Significant difference in the percentage of nonzero coherence and GC values at P < 0.001 between age groups. The SO, Delta, Theta, Alpha, Beta, and Gamma refer to the EEG power in these frequency bands. The PE is the entropy values derived from the EEG. MOSSA, maintenance of a surgical state of anesthesia; GC, Granger-causality; SO, slow oscillation; PE, permutation entropy.

Download Full Size | PDF

For each state, $C{O_{SO - Hb}}$ and $C{O_{PE - Hb}}$ values of adults are significantly higher than those of children in wakefulness and recovery (p < 0.05), and $C{O_{SO - HbO}}$ and $C{O_{delta - HbO}}$ values of adults are significantly higher than those of children in wakefulness (p < 0.05). Besides, the coherences between theta, alpha and gamma, and hemodynamic activities of children are significantly higher than those of adults during MOSSA (p < 0.01). For each age group, there are significant increases from wakefulness to MOSSA and significant decreases from MOSSA to recovery (p < 0.01) in most coherence indices (see Table S12 and S13), especially from wakefulness to MOSSA. The results indicate that there is age-dependent coherence between ${P_{theta}}$, ${P_{alpha}}$ and ${P_{gamma}}$, and hemodynamic oscillations in MOSSA. This neurovascular coupling in the frequency domain for children is stronger than that of adults.

Finally, we analyzed causality from EEG activities to hemodynamic responses in both age groups during the three states by using the GC measure. The heatmaps of the GC from EEG activities (${P_{SO}}$, ${P_{delta}}$, ${P_{theta}}$, ${P_{alpha}}$, ${P_{beta}}$, ${P_{gamma}}$ and PE) to hemodynamic activities are shown in Figure 6(C) and (D). The mean ± SD for GC across age groups during various states is shown in Tables S14 and S15 (in Supplement 1). The significance tests for all of these indices between age groups in each state are shown in Table S16 and among various states in each age group are shown in Table S17 and S18 (in Supplement 1).

For each state, we find no significant differences between the two age groups for all GC indices, except for $G{C_{theta \to HbO}}$, which means that the causality connection from EEG activities to hemodynamic responses are not age-dependent. For each age group, we found that the all GC indices decreased from wakefulness to MOSSA, and increased from MOSSA to recovery. There are significant differences in GC indices from EEG powers to hemodynamic activities between MOSSA and the other two states for adults (p < 0.05), but not for children. The $G{C_{SO \to Hb}}$ and $G{C_{beta \to Hb}}$ significantly decreased from wakefulness to MOSSA, and increased from MOSSA to recovery in two age groups (p < 0.05). There are significant differences in GC indices from PE to hemodynamic activities between wakefulness and MOSSA, except for $G{C_{PE \to HbO}}$ in children. This indicates that GC can better distinguish the anesthesia states in adults than in children. $G{C_{SO \to Hb}}$ and $G{C_{beta \to Hb}}$ could be used as a new DoA index for both children and adults.

4. Discussion

Investigating the age-dependent neuronal and hemodynamic changes during general anesthesia is of critical importance for revealing the effect mechanism of anesthetic drugs across ages and for developing a stable age-related DoA index. In this study, we used simultaneous EEG and fNIRS systems to obtain the EEG, Δ[Hb], Δ[HbO] and Δ[HbT] during general anesthesia in both children and adults. The results suggest that the neuronal activities, hemodynamic activities and neurovascular coupling are age-dependent.

4.1 Frontal EEG power is age-dependent but PE shows no age difference between 6-12 years old children and adults

The age-dependent EEG spectrum during MOSSA has been demonstrated in previous studies [9,50,51]. However, whether this is the case throughout the time course is unknown. The underlying maturation of the cerebral cortex, the rates of synaptogenesis, white-matter tract development, and thalamocortical and cortico-thalamic connections, may cause the variegated cortical rhythms from infancy, childhood, to adulthood [52–54]. In this study, we found that the EEG powers in children (6-12 years old) during wakefulness and MOSSA were stronger than those in adults. This matches the brain development process, such as in synaptic pruning and synaptic refinement.

PE has been proved to be an effective method for DoA monitoring [34–36]. However, most studies have focused on adults. In recent years, Kim et al. [36] have shown that the adjusted PE has a higher correlation than the BIS in children 3 to 15 years old during sevoflurane anesthesia. Kreuzer et al. [34] found age-related changes in the PE from adulthood to old age during sevoflurane anesthesia in the MOSSA state . In this study, we found PE to be as effective as BIS in distinguishing anesthesia states in both children and adults. There were no significant differences between children and adults in the states of either wakefulness or MOSSA. This indicates that the PE is an age-insensitive index from children to adults.

A recent study has shown that PE increases in a wide frequency band (0.5-30 Hz) from adults to the elderly during MOSSA [34]. They suggested that the ordinal EEG pattern probably reflects the effect on the 1/f slope in the EEG spectrum. As shown in Figures S1(A) and (B) (representing the states of wakefulness and MOSSA), we found that the 1/f slopes in the EEG spectra of children and adults were similar. However, the 1/f slope of the EEG spectra in the recovery state varies between children and adults. This is consistent with the finding that there are significant differences between children’s and adults’ PE indices in recovery (p < 0.05). Therefore, we hypothesize that the EEG spectrum’s 1/f slope, not the amplitude or frequency itself, has a strong correlation with PE. Also, the excitatory synaptic strength in the prefrontal cortex peaks around 5-years-old, and then plateaus at a lower level near 18-years-old [54]. We hypothesize that this excitatory synaptic strength change may influence the EEG oscillations’ ordinal irregularity characteristics.

In the present study, we found no alpha rhythm observed during wakefulness (and/or recovery) in adults. Usually, the amplitude of alpha EEG is the highest in the state of non-task wakefulness and closed eyes (i.e., resting state) [55]. The patients were in a nervous state, especially the child before anesthesia. It may not be the same as that in the normal resting state. Recovery is a complex process. In our previous study [30], we found four different EEG spectrum patterns during the recovery process. They are pattern I: alpha loss, delta persistent; pattern II: alpha and delta wave loss; pattern III: alpha and delta persistent, and pattern IV: only delta persistent. The pattern I and pattern III are age-related. Also, it was suggested that the recovery of consciousness is not a mirror image of LOC [2,56,57]. Therefore, it is not a special phenomenon that alpha oscillation in wakefulness and recovery has not been observed. For awake state, it may depend on the state of the patient (relaxed or nervous) and whether the eyes are closed or not. For recovery, the spectrogram of EEG may be influenced by the drug used, age, and whether the eyes are closed or not.

In our study, compared to adults, there was no significant variation in error bars and interquartile ranges for children, except for some EEG power indices (Alpha/Delta) and (Beta + Alpha)/(Delta + Theta) in MOSSA. We think that this may be due to the small sample size. It is also possible that developmental changes do not cause much fluctuation in the indicators. Moreover, due to the small sample size, we did not make more detailed age segmentation, such as 6-9 years and 10-12 years, so the indicators changing with age could not be observed.

4.2 Both neuronal and physiological effects should be considered in interpreting hemodynamic changes

FNIRS detects the hemodynamic changes in the cerebral cortex which is an indirect reflection of neuronal activity through neurovascular coupling. FNIRS can also reflect physiological activity through changes in blood pressure and cerebral blood flow. In this study, we analyzed the indices (i.e., Δ[Hb] and Δ[HbO]) from fNIRS in the frequency band below 0.1 Hz. We found that the Δ[Hb] decreased and Δ[HbO] increased after propofol-induced anesthesia, in both children and adults (see Figures 4(C) and (D)).

Several studies have investigated Δ[Hb] and Δ[HbO] changes during general anesthesia without reaching conclusive findings. Curtin et al. [7] found that Δ[Hb] decreased and Δ[HbO] increased after bolus infusions of propofol during outpatient elective colonoscopy. This finding accords with our results somewhat. In contrast, Leon-Dominguez et al. [6] found that the Δ[Hb] increases during suppression, but increases during emergence with patients undergoing propofol- and sevoflurane-induced general anesthesia during coloproctology surgery. In our previous study [58], we found that both the Δ[Hb] and the Δ[HbO] increased after LOC and decreased after ROC in patients undergoing propofol-induced and sevoflurane-maintained general anesthesia. These inconsistent findings indicate anesthetic drug-induced complex hemodynamic change. It has been suggested that propofol decreases the regional CBF, cerebral metabolic rate of oxygen (CMRO2) and cerebral blood volume (CBV) [59]. Meanwhile, propofol also lead to systemic vasodilation and reduces systemic mean arterial pressure, and increases blood flow to peripheral areas [60]. Additionally, it has been found that the changes in cerebral hemodynamics caused by anesthetic are spatially non-uniform [61], and have intra-subject variability [62]. All these factors may cause the different conclusion in different studies. Therefore, interpreting these changes and employing them to monitor the cerebral response to anesthetic drugs warrants consideration of the physiological effects and the influence of systemic, superficial, and other global effects in fNIRS measurement [63].

Further, in this study, we found a significant difference between age groups in Δ[Hb] during MOSSA and in Δ[HbT] while awake. It has been suggested that the deoxy-hemoglobin fluctuations have a strong correlation with the BOLD responses in fMRI [64]. This could be a reflection of the effects of neuronal activity changes. Therefore, we hypothesized that the age-related neuronal activities, such as EEG power, may cause the age difference in Δ[Hb] during MOSSA. The change in total oxyhemoglobin can reflect the variations in regional cerebral blood volume. Moreover, a previous study [65] has shown that brain development causes the increased blood volume, capillary and venous blood flow, and mean arterial pressure (limited by cerebral auto-regulation) in infants. Wu et al. [66] have studied CBF in a cohort of children (7 months-17 years old) and adults (19-61 years old) with the fMRI technique. They found an increase in peak velocity in several major arteries of the brain until about 7 years, and a decrease until 18-20 years old. We hypothesize that these age-related vascular changes are the reason for the variation in Δ[HbT] across ages while awake.

4.3 Neurovascular coupling indicators offer the potential for designing age-independent DoA indices

Although the EEG is widely used, there are some advantages of fNIRS in monitoring DoA. First, EEG can produce false alarms when there is electromagnetic interference [67]. Second, most EEG-based DoA systems, such as BIS, are suitable only for the monitoring of GABAergic anesthetics (i.e., propofol and sevoflurane), but are not be suitable for the monitoring of the anesthesia drugs of N-methyl-D-aspartate (NMDA) receptor (i.e., ketamine) [68]. However, the Δ[Hb] and Δ[HbO] obtained by fNIRS can be used to reflect CBF and effectively monitor DoA under ketamine drug anesthesia. Third, the hemodynamic signal of Δ[Hb], Δ[HbO] and Δ[HbT] derived from fNIRS could reflect the change of CBF and cerebral oxygen metabolism of the brain, which are complementary to the EEG based DoA monitoring for surgical patients [69]. Finally, the multimodal of EEG-fNIRS measurement could investigate the neurovascular coupling between neuronal activity and hemodynamic oscillations, which is better than unimodal measurement [45,70,71]. In present study, we used the correlation, coherence, and GC to measure the neurovascular coupling between neuronal activity and hemodynamics from the perspective of temporal, frequency, and causality domains, respectively. We found that the neurovascular coupling indices could distinguish different anesthesia states and some of them are age-dependent (coherences between ${P_{delta}}$, ${P_{theta}}$, and ${P_{gamma}}$, and hemodynamic activities). These metrics could be used to reveal the relationship between neuronal activity and cerebral metabolism/vascular changes under the action of anesthetics, which supplies a new perspective for studying the age-related effect mechanism of anesthesia. Therefore, it is valuable to employ the multimodal tool to evaluate effects of anesthetic drugs at different ages.

In this study, we analyzed the coupling between the EEG indices and the hemodynamic responses from different perspectives. For the correlation, in contrast to the findings of Leon-Dominguez et al. [6], we found a positive correlation between the Δ[Hb] index and EEG indices. We also found that the correlation between PE and hemodynamic responses in adults were higher than those in children. This suggests that the neurovascular coupling is high, relative to the stages of brain development.

For the coherence, we found that the coherence increased in MOSSA and then decreased in recovery for all coherence indices. Also, there was no significant difference between different age groups for most of the indices. This implies that the coherence between EEG and hemodynamic indices should be a valuable age-independent DoA index for clinical setting. While, in all these coherence indices, we found the coherences between theta, alpha and gamma, and hemodynamic activities have significant differences between children and adults in MOSSA. This may be caused by the distinct power difference in theta and alpha for these two age groups (see in Figure S1). Our results suggested that the coherence supplies a new tool to investigate the mechanism of neurovascular coupling during general anesthesia.

Further, we investigate the GC from EEG indices to hemodynamic responses. The results of GC analysis showed that the causality weakened after LOC and strengthened after recovery, in both age groups. This indicates that the dependencies between neuronal activities and hemodynamic responses decreased after LOC. Also, there were no significant age differences between the children’s GC and those of adults. This means that the couplings between neuronal activities and hemodynamic responses were not affected by age. This information is essential to designing a next-generation DoA monitor.

Multimodal monitoring could provide vital information for understanding the effect of anesthetics and guaranteeing the safety of patients. Various studies have investigated the relationship between neuronal activity-based indices and hemodynamics during general anesthesia [72], ischemic stroke [73], and sleep [74]. All these studies suggested that multimodal physiological parameters supply comprehensive investigation tool for understanding the complex system of the brain. The measures employed in this study provide novel approaches for investigating the effect mechanism of anesthetics in multimodal monitoring.

5. Limitations

This study has the following limitations. First, the sample sizes, for both children and adults, were relatively small. Second, due to clinical requirements, drug concentrations were not uniform across patients during MOSSA. Thus, it is possible that differences in EEG characteristics may have been due to drug concentration, rather than age. However, our analysis showed no significant differences in MAC levels for the patients that we analyzed. Third, we only analyzed the fNIRS signals on the same side of the forehead as the BIS, so the spatial processing and coupling analysis between modalities was absent. Fourth, unlike some other fNIRS systems, such as CW6 (TechEn Inc., MA, U.S.A) and ETG-4000 NIRS system (Hitachi Medical Corporation), the fNIRS system we used could only output the Δ[Hb] and Δ[HbO] indices from the devices but not the raw light intensity values. So, we cannot do any correction for calculating the Δ[Hb] and Δ[HbO] by using the age-adaptive DPF. Fifth, the system only outputs Δ[Hb] and Δ[HbO] information of the source-detector distance at 4 cm, but not the information of two distances (i.e., 4 cm and 3 cm). So, our Δ[Hb] and Δ[HbO] contain the oxygen metabolism information of scalp and cerebral cortex. We cannot remove the influence of skin vasculature by using the short distance regression due to the lack of information of Δ[Hb] and Δ[HbO] in 3 cm [19]. Finally, since the physiology information was not concurrently recorded in our experiment, we cannot obtain the exact effect of these factors on hemodynamic change. Besides, the non-neural actors, i.e., fluctuations in blood oxygenation and flow, also have a great effect on hemodynamic activity [75]. This influence should be considered in future study.

6. Conclusion

The EEG-derived indices (power, PE) and fNIRS-derived indices (Δ[Hb] and Δ[HbO]), as well as the coupling (correlation, coherence and GC) between modalities show varying degrees of age-dependency. The PE, Δ[Hb], coherence and GC indices can better distinguish brain states and different age groups. The multimodal EEG-fNIRS dynamics analysis methods in this study have potential to monitor the effect of anesthesia on different age groups during general anesthesia.