Changes in oxygenation levels precede changes in amplitude of the EEG in premature infants

Changes in oxygenation levels precede changes in amplitude of the EEG in premature infants

Alexander Caicedo^a,b, Liesbeth Thewissen^c,d, Anne Smits^e, Gunnar Naulaers^c,d, Karel Allegaert^c,d, Sabine Van Huffel^a,b

a Department of Electrical Engineering, STADIUS-ESAT, KU Leuven, Belgium

b iMinds, Medical IT

c Department of development and regeneration, KU Leuven

d Department of Neonatology, University Hospitals Leuven

e Department of Pediatrics, University Hospitals Leuven, Belgium

Abstract Brain function is supported by an appropriate balance between the metabolic demand and the supply of nutrients and oxygen. However, the physiological principles behind the regulation of brain metabolism and demand in premature infants are unknown. Some studies found that changes in hemodynamic variables in this population precede changes in EEG activity; however, these studies only used descriptive statistics. This paper describes the relationship between changes in cerebral oxygenation, assessed by means of near-infrared spectroscopy (NIRS), and changes in EEG, using mathematical methods taken from information dynamics. In a cohort of 35 neonates subjected to sedation by propofol, we quantified the direction of information transfer between brain oxygenation and EEG. The results obtained indicate that, as reported in other studies, changes in NIRS are likely to precede changes in EEG activity.

1 Introduction

Brain metabolism is supported by adequate cerebral hemodynamic regulation, which provides the necessary substrates and is reflected in appropriate brain functioning. Oxygen is one of the most important substrates that is needed in order to meet the energy demand of the brain. For monitoring purposes, brain oxygenation levels can be assessed by means of near-infrared spectroscopy (NIRS), while brain function can be assessed by means of EEG [1]. Several studies have shown that EEG is a very good predictor of early neonatal outcome, especially in asphyxiated infants [2]; however, its prognostic value is decreased in cooled neonates [3]. Changes in brain oxygenation probably relate to changes in

brain function. Consequently, combining the information provided by brain oxygenation levels and brain function enables to assess the coupling of the cerebral metabolic demand and oxygen delivery to the brain. This coupling was shown to be lost in case of pathological conditions such as hypoxic-ischemic encephalopathy [4, 5]. However, these studies are only a description of observations and lack metrics that are able to adequately measure the link between the dynamics of cerebral hemodynamics regulation and EEG.

In this paper we attempt to identify whether changes in brain oxygenation precede changes in EEG in premature infants, as shown in [5], but using a quantitative method based on information dynamics instead of descriptive statistics.

2 Methods

Data were acquired from 35 neonates undergoing an intubation procedure in the Neonatal Intensive Care Unit. The data were recorded as part of a study to identify the optimal dose of propofol for procedural sedation in neonates (ClinicalTrials.gov NCT01621373). Concomitant measurements of brain oxygenation, measured by NIRS (INVOS 5100, Covidien, neonatal probe) with the optode located left frontoparietal, and raw EEG, from channels C3-C4, were obtained from the patients (Olympic CFM 6000, Natus). NIRS signals were measured at 1Hz, while EEG signals were acquired at 100Hz. The measurements were carried out before administration of propofol and lasted up to 12 hours. In this study we analysed data from 2 min before up to 10 hours after protocol administration in order to have a homogeneous dataset. All data were pre- processed by detecting artefacts manually and replacing them by NaN (Not a Number) in the data stream. For the EEG measurements, segments of data with an impedance higher than 10KΩ were identified and replaced by NaN, since the measurement is unreliable.

Due to the different temporal characteristics between the NIRS signals and EEG, we used a continuous estimate of the power contained in the EEG signal, by computing a running root mean squared (RMS) value of the EEG. For this purpose we used a window length of 60 s, with an overlapping of 59 s, producing one new value every second. In this way both signals, NIRS and the power contained in the EEG, have a common temporal scale and sampling frequency. The 60-s window for the RMS computation was selected since the spectral decomposition of the NIRS signal indicates that most of its power is located below 0.16 Hz. Therefore, to keep the power of this component, a window of around 1-min length was needed; shorter windows will introduce components at a higher frequency, while longer windows will filter components that we would like to keep. Figure 1 is a representative sample of the recordings.

In order to identify the directionality of the relation between brain oxygenation levels and EEG power, we use transfer entropy. Transfer entropy is a measure for

the amount of information transferred from one signal (process) to another [6]. A transfer entropy equal to zero indicates no transfer of information in the established direction, whilst any other value indicates a link between the processes. Higher values of transfer entropy indicate a higher transfer of information and, hence, a stronger directional coupling [7]. The transfer entropy computes the rate of information transfer between two different signals, namely X and Y. This computation starts from an autoregressive model between X and Y, using L delays in X and K delays in Y as follows:





 ^L_{r r n r} ^K_{q p n p}

n a x b y

y _{1 1 1}

where ar and bq represent the coefficients for this model, and can be simplified in the following form y_n a^{   L}x_n^L b^{   K} y_n^K , where the upper script (.) represents a set of ensemble, K or L, samples that form a vector of the past instances of the respective signals. Transfer entropy is then computed as follows:

   

  













  _K

n n

L n K n L n

n K n n Y

X p y y

x y y x p

y y p T

1 1 1

log , ,

,

where X and Y represent the two time series,





K n

n y x

y

p _1, , represents the joint probability distribution, and





K n

n y x

y

p _1 , and p





represents the conditional probability distributions.

In order to identify the direction of information transfer, and whether changes in oxygenation levels precede the changes in EEG activity, we computed

EEG

TNIRS_ and TEEG_NIRS .

To compute the transfer entropy values we proceed as follows:

1. The systemic and EEG signals were preprocessed as indicated before.

2. The time at which the propofol was applied was indicated. The signals were segmented from 2 min before propofol administration up to 10 hours.

3. The transfer entropy values were computed for the first 2 min prior to propofol administration in order to generate a reference value for the baseline.

4. The remaining part of the signal was segmented in non-overlapping windows of 5 min. For each window, TNIRS_EEG and TEEG_NIRS

were computed.

5. For each window the method evaluates the order of the model, specifically K and L, we have imposed as the maximum model order 20 samples. This is done to guarantee that there are enough samples to compute the probability distributions.

6. The results were analyzed using a Kruskal-Wallis test in order to indicate whether the differences between the two values are significant.

3 Results

Figure 2 shows the values of TENIRSEEG and TEEEGNIRS for the 35 neonates, during the 10 hours of measurements. The figure presents the median as a solid line, and the 25-75 percentiles as a shaded area; also presented are the values for TENIRSEEG

in blue and the values for TEEEGNIRS in red.

For the statistical analysis, all data were collected in two variables, each one representing each TE measurement. The values were dispersed as follows:

TENIRSEEG [0.0036, 0.0062, 0.0202] and TEEEGNIRS [0.0007, 0.0019, 0. 0128];

where the numbers within square brackets represent the 25, 50 and 75 percentiles respectively. Results from the Kruskal-Wallis test indicated that both TENIRSEEG

values were significantly higher than TEEEGNIRS values, with p<0.001.

4 Discussion and Conclusions

In this study, we found that the transfer of information, measured by means of transfer entropy, is larger in the direction of NIRSEEG than in the direction from EEGNIRS. This indicates that the changes in NIRS are likely to precede the changes in EEG. These results are in agreement with the findings of Roche- Labarbe et al., who describe that the changes in NIRS precede the changes observed in EEG [5]. In their study they investigated whether bursts of EEG activity were coupled to a hemodynamic response in premature infants. They found that changes in hemodynamic parameters were sometimes observed a few seconds before onset of the EEG activity. They hypothesize that these observations might be due mainly to the fact that the neural activity might have started before it was observed in the EEG. The cause for this phenomenon can be multifactorial and may be due to the fact that EEG measures only the superficial changes in cerebral hemodynamics and not more internal layers, where many processes might occur and not be reflected in the EEG; this is further discussed in [5]. We would like to stress that in our study we only indicate that, according to the values provided by transfer entropy, changes in NIRS are more likely to precede the changes in EEG, than otherwise. To the best of our knowledge, this is the first study that quantifies the direction of information transfer between NIRS and EEG in general.

Since the transfer of information EEGNIRS is not zero, this indicates that there is a flow of information in this direction. This can be attributed to the feedback mechanisms that are in charge of the regulation of brain hemodynamics, and which are in charge of delivering nutrients and oxygen to meet the metabolic

demand. These results indicate that the analysis of EEG-NIRS, using transfer entropy, can be used to assess the status of these regulatory mechanisms in the premature brain.

Additionally, we expected some effect of the application of propofol in the coupling between the signals. However, this was not observed in the present results. We hypothesized that this might be due to the fact that propofol not only affects the hemodynamic regulatory mechanisms due to its vasodilator action, but also produces a depression in the EEG, via other mechanisms. These two responses might have been equally affected, producing no change in the transfer entropy values.

It is important to note that transfer entropy assumes that, in the selected window, the processes under analysis are stationary. In addition, the processes are assumed to be linearly related and to have a Gaussian distribution. We are aware that these conditions might not be fulfilled in the selected window of 5 min.

However, even though by selecting shorter windows these conditions are more likely to be met, the number of samples in the window under analysis then becomes too small to provide reliable probability distribution, as well as estimation of model parameters which are, afterwards. used for the estimation of the transfer entropy values.

In conclusion, the results presented in this study confirm the observations of Roche-Labarbe et al., by providing a quantification for the directionality of information transfer between NIRS and EEG measurements. The physiological mechanism of this phenomena is still unclear and requires more studies.

Acknowledgments Alexander Caicedo is a postdoctoral fellow of the research foundation Flan- ders (FWO). Bijzonder Onderzoeksfonds KU Leuven (BOF). This research was also supported by: Center of Excellence (CoE) #: PFV/10/002 (OPTEC). Fonds voor Wetenschappelijk Onderzoek-Vlaanderen (FWO), project #: G.0427.10N (Integrated EEG-fMRI), G.0108.11 (Compressed Sensing), G.0869.12N (Tumor imaging), G.0A5513N (Deep brain stimulation).

Agentschap voor Innovatie door Wetenschap en Technologie (IWT), project #: TBM 080658- MRI (EEG-fMRI), TBM 110697-NeoGuard. iMinds Medical Information Technologies..

Dotatie-Strategisch basis onderzoek (SBO- 2015). ICON: NXT_Sleep. Belgian Federal Science Policy Office. IUAP #P7/19/ (DYSCO, `Dynamical systems, control and optimization', 2012- 2017). Belgian Foreign Affairs-Development Cooperation. VLIR UOS programs (2013-2019).

EU: European Union's Seventh Framework Programme (FP7/2007-2013): EU MC ITN TRANS- ACT 2012, #316679, ERASMUS EQR: Community service engineer, #539642-LLP-1-2013.

Other EU: INTERREG IVB NWE programme #RECAP 209G. European Research Council:

ERC Advanced Grant, #339804 BIOTENSORS This paper reflects only the authors' views and the Union is not liable for any use that may be made of the contained information.

References

1.Prior PF, Maynard DE (1986). Monitoring cerebral function. Long-term recordings of cerebral electrical activity and evoked potentials. Elsevier, 1-441.

2.Hallberg B, Grossmann K, Bartocci M, et al (2010). The prognostic value of early aEEG in asphyxiated infants undergoing systemic hypothermia treatment. Acta Paediatr, 99.4: 531-536.

3.Ancora G, Maranella E, Grandi S, et al (2013). Early predictors of short term neurodevelopmental outcome in asphyxiated cooled infants. A combined brain amplitude integrated electroencephalography and near infrared spectroscopy study. Brain Dev, 35(1):26- 31.

4.Pichler G, Avian A, Binder C, et al (2013). aEEG and NIRS during transition and resuscitation after birth: Promising additional tools; an observational study. Resuscitation 84(7):974-978.

5.Roche-Labarbe N, Wallois F, Ponchel E, et al (2007). Coupled oxygenation oscillation measured by NIRS and intermittent cerebral activation on EEG in premature infants.

Neuroimage, 36.3:718-727

6.Schreiber T (2000) Measuring information transfer. Phys Rev Lett 85: 461. doi:

10.1103/physrevlett.85.461

7.Faes L, Marinazzo D, Montalto A, et al (2014). Lag-specific transfer entropy as a tool to assess cardiovascular and cardiorespiratory information transfer. IEEE Trans Biomed Eng, 61(10):

2556-2568.

Fig. 1 Representative set of measurements for one subject. The rScO2, preprocessed EEG and raw EEG measurements are shown.

Fig. 2 Transfer entropy values. The solid line represents the median over the different epochs across time, the shaded areas represent the 25% and 75% percentiles for the TE values across the group, the blue values represent TE_NIRSEEG, while the red ones represent

NIRS

TEEEG .