Automated analysis of multi-channel EEG in preterm infants

doi:10.1016/j.clinph.2014.11.024

Clinical Neurophysiology

Volume 126, Issue 9, September 2015, Pages 1692-1702

https://doi.org/10.1016/j.clinph.2014.11.024 Get rights and content

Highlights

•
Automatic detection of burst and interburst periods in multi-channel preterm EEG.
•
EEG marked by 3 independent observers, detailed interobserver analysis.
•
Mathematical features of the EEG are calculated and correlated with gestational age.

Abstract

Objective

To develop and validate two automatic methods for the detection of burst and interburst periods in preterm eight-channel electroencephalographs (EEG). To perform a detailed analysis of interobserver agreement on burst and interburst periods and use this as a benchmark for the performance of the automatic methods. To examine mathematical features of the EEG signal and their potential correlation with gestational age.

Methods

Multi-channel EEG from 36 infants, born at less than 30 weeks gestation was utilised, with a 10 min artifact-free epoch selected for each subject. Three independent expert observers annotated all EEG activity bursts in the dataset. Two automatic algorithms for burst/interburst detection were applied to the EEG data and their performances were analysed and compared with interobserver agreement. A total of 12 mathematical features of the EEG signal were calculated and correlated with gestational age.

Results

The mean interobserver agreement was found to be 77% while mean algorithm/observer agreement was 81%. Six of the mathematical features calculated (spectral entropy, Higuchi fractal dimension, spectral edge frequency, variance, extrema median and Hilberts transform amplitude) were found to have significant correlation with gestational age.

Conclusions

Automatic detection of burst/interburst periods has been performed in multi-channel EEG of 36 preterm infants. The algorithm agreement with expert observers is found to be on a par with interobserver agreement. Mathematical features of EEG have been calculated which show significant correlation with gestational age.

Significance

Automatic analysis of preterm multi-channel EEG is possible. The methods described here have the potential to be incorporated into a fully automatic system to quantitatively assess brain maturity from preterm EEG.

Introduction

The survival rate of very preterm infants is increasing (Costeloe et al., 2012) and it is recognised that EEG is an extremely useful and non-invasive tool for monitoring neurological well being and for the prediction of neurological outcome (LeBihannic et al., 2012, Biagioni et al., 1994, Biagioni et al., 1996, Holmes and Lombroso, 1993, Marret et al., 1997, Hayashi-Kurahashi et al., 2012). With recent advances in technology it is now much more feasible to use preterm EEG monitoring for longer periods of time (Schumacher et al., 2011). The very preterm EEG typically exhibits a discontinuous pattern of high voltage bursts of activity interspersed with low voltage periods known as interburst intervals (IBIs). As the infant matures, IBIs become gradually shorter, bursts become longer and by the time the infant reaches full term the EEG has a continuous pattern (Vecchierini et al., 2007). The appropriate length of IBIs at particular gestational ages varies widely according to different reports (Hahn et al., 1989, Biagioni et al., 1994, Selton et al., 2000, Hayakawa et al., 2001, Vecchierini et al., 2003, Vecchierini et al., 2007), however it has been shown that long IBIs (>1 min) are abnormal at any gestational age (Vecchierini et al., 2007).

Neurophysiologists generally assess brain maturity in the preterm EEG by measuring the duration of burst and interburst intervals and assessing age specific patterns. Since the opinion of the neurophysiologist is subjective and only uses selected representative periods of EEG for analysis, this task results in an output which is not reproducible or quantitative. For this reason, some recent studies have investigated automated methods of detecting bursts and IBIs in preterm EEG (Palmu et al., 2010, Jennekens et al., 2011, Koolen et al., 2014). In this way, measurements can be made automatically and virtually instantaneously.

Previous methods of automated burst detection have been developed but have been limited to either single channel recordings or to very small numbers of subjects. Palmu et al. (2010) use data from 18 preterm infants with single channel EEG. Jennekens et al. (2011) use a database of just four infants, and Koolen et al. (2014) just 10, including seven with confounding conditions such as suspected seizures, mild asphyxia or intracranial haemorrhage. Although these studies provided promising results, larger databases are required to determine whether they are reproducible. Furthermore, these studies lack detailed comparisons with multiple observers and interobserver analysis.

The aim of this study was to develop and validate two automatic methods for the detection of bursting patterns in eight-channel preterm EEG. Furthermore, we perform a detailed analysis of inter-observer agreement between three expert observers and use this as a benchmark for the performance of the automatic methods. Finally, we examine mathematical features of the EEG signal and their potential correlation with gestational age.

Section snippets

Material

The data used consists of eight-channel EEG recordings on 36 preterm infants born at less than 30 weeks gestation. Fig. 1 shows the distribution of the subjects across gestational ages.

EEG recording was commenced at a median age of 6 h and 15 min (minimum 3 h, maximum 23 h 11 min). The recordings lasted between 7 h 55 min and 71 h 14 min (median 46 h 57 min). All EEGs were recorded over eight channels as follows: F4–C4, C4–O2, F3–C3, C3–O1, T4–C4, C4–Cz, Cz–C3, C3–T3. Sampling rate was 256 Hz for 32 of the

Burst detection

In this section the results of the burst detection algorithm are shown and a comparison between the ‘fixed threshold’ method and the ‘variable threshold’ method is made.

Fig. 2 shows the performance of the two automatic algorithms based on overlap in regions where observers were unanimous. The first group of boxplots, shown in green, give results for the ‘fixed threshold’ method, using various fixed thresholds, and the second group (in blue) show the results for the ‘variable threshold’ method

Burst detection

In this study, a method for burst detection in multi-channel EEG of 36 preterm infants at less than 30 weeks gestation has been presented. The NLEO operator was utilised to measure activity in the EEG signal and two different methods of thresholding the NLEO output were implemented. In our dataset of 36 preterms we were unable to detect a significant difference between the output of the two methods (fixed threshold or variable threshold on NLEO). A paired t-test on the best fixed-threshold

Acknowledgements

This research was supported by a Science Foundation Ireland Research Centre Award (INFANT – 12/RC/2272).

Conflict of interest: None of the authors have potential conflicts of interest to be disclosed.

References (25)

R. Agarwal et al.
Automatic EEG analysis during long-term monitoring in the ICU
Electroencephalogr Clin Neurophysiol
(1998)
E. Biagioni et al.
Background EEG activity in preterm infants: correlation of outcome with selected maturational features
Electroencephalogr Clin Neurophysiol
(1994)
W. Deburchgraeve et al.
Automated neonatal seizure detection mimicking a human observer reading EEG
Clin Neurophysiol
(2008)
J.S. Hahn et al.
Interburst interval measurements in the EEGs of premature infants with normal neurological outcome
Electroencephalogr Clin Neurophysiol
(1989)
T. Higuchi
Approach to an irregular time series on the basis of the fractal theory
Phys D Nonlinear Phenomena
(1988)
K. Holthausen et al.
Brain dysmaturity index for automatic detection of high-risk infants
Pediatric Neurol
(2000)
N. Koolen et al.
Line length as a robust method to detect high-activity events: Automated burst detection in premature EEG recordings
Clin Neurophysiol
(2014)
S. Marret et al.
Prognostic value of neonatal electroencephalography in premature newborns less than 33 weeks of gestational age
Electroencephalogr Clin Neurophysiol
(1997)
D. Selton et al.
Normal EEG in very premature infants: reference criteria
Clin Neurophysiol
(2000)
M.J.A.M. van Putten
The revised brain symmetry index
Clin Neurophysiol
(2007)

M.-F. Vecchierini et al.

Normal EEG of premature infants born between 24 and 30 weeks gestational age: terminology, definitions and maturation aspects

Clin Neurophysiol

(2007)

M.-F. Vecchierini et al.

EEG patterns in 10 extreme premature neonates with normal neurological outcome: qualitative and quantitative data

Brain Dev

(2003)

Cited by (31)

Quantitative analysis of high-frequency activity in neonatal EEG
2023, Computers in Biology and Medicine
To determine the presence and potential utility of independent high-frequency activity recorded from scalp electrodes in the electroencephalogram (EEG) of newborns.
We compare interburst intervals and continuous activity at different frequencies for EEGs retrospectively recorded at 256 Hz from 4 newborn groups: 1) 36 preterms (<32 weeks' gestational age, GA); 2) 12 preterms (32–37 weeks' GA); 3) 91 healthy full terms; 4) 15 full terms with hypoxic-ischemic encephalopathy (HIE). At 4 standard frequency bands (delta, 0.5–3 Hz; theta, 3–8 Hz; alpha, 8–15 Hz; beta, 15–30 Hz) and 3 higher-frequency bands (gamma1, 30–48 Hz; gamma2, 52–99 Hz; gamma3, 107–127 Hz), we compared power spectral densities (PSDs), quantitative features, and machine learning model performance. Feature selection and further machine learning methods were performed on one cohort.
We found significant (P < 0.01) differences in PSDs, quantitative analysis, and machine learning modelling at the higher-frequency bands. Machine learning models using only high-frequency features performed best in preterm groups 1 and 2 with a median (95% confidence interval, CI) Matthews correlation coefficient (MCC) of 0.71 (0.12–0.88) and 0.66 (0.36–0.76) respectively. Interburst interval-detector models using both high- and standard-bandwidths produced the highest median MCCs in all four groups.
High-frequency features were largely independent of standard-bandwidth features, with only 11/84 (13.1%) of correlations statistically significant. Feature selection methods produced 7 to 9 high-frequency features in the top 20 feature set.
This is the first study to identify independent high-frequency activity in newborn EEG using in-depth quantitative analysis. Expanding the EEG bandwidths of analysis has the potential to improve both quantitative and machine-learning analysis, particularly in preterm EEG.
CARFS<sup>7</sup>: A guide and proforma for reading a preterm neonate's EEG
2022, Neurophysiologie Clinique
Citation Excerpt :
The role of the CARFS7 proforma in prognostication must be evaluated in larger cohorts of preterm babies. Automated analysis of preterm multichannel EEG [47] has the potential to be available in future in clinical practice. Our proforma, with a scoring system for the important EEG components, may provide supportive data to engineers and neuroscientists in their quest for development of efficient, practical automated programs for interpretation and prognostication based on preterm EEGs.
The important role of the EEG in preterm and term babies in investigating brain function and seizures, predicting outcomes, evaluating therapeutic interventions and decision-making is being increasingly acknowledged. Development of the brain in the last trimester of pregnancy results in rapid changes in the EEG patterns in this period. Acquiring and interpreting the EEG of a preterm baby can be challenging. The aim of this study was to develop a proforma titled CARFS⁷ (Continuity, Amplitude, Reactivity, Frequency, Synchrony, Symmetry, Sleep, Sharps, Shapes, Size and Seizures) to enable neurologists to read EEGs of premature babies with greater confidence, ease and accuracy and produce a report more easily repeatable and homogenous among operators.
The CARFS⁷proforma was developed based on a literature review and the personal experience of the authors. The parameters of the EEG evaluated and scored in the proforma are Continuity, Amplitude, Reactivity/Variability, Frequency, Synchrony, Symmetry, Sleep, Sharps, Shapes/Patterns, Size and Seizures. We also assessed the interrater reliability of the proposed scoring system incorporated in the proforma.
CARFS⁷ proforma incorporates a number of parameters that help evaluate the preterm EEG. The interrater reliability of the proposed scoring system in the CARFS⁷proforma was high.
CARFS⁷ is a user friendly proforma for reading EEGs in the preterm infant. Interrater reliability using Cohen's k shows high agreement between two child neurologists who independently rated the EEGs of 25 premature babies using this proforma. CARFS⁷ has the potential to provide, accurate, reproducible and valuable information on brain function in the preterm infant in clinical practice.
Back to basics: the neuronal substrates and mechanisms that underlie the electroencephalogram in premature neonates
2021, Neurophysiologie Clinique
Electroencephalography is the only clinically available technique that can address the premature neonate normal and pathological functional development week after week. The changes in the electroencephalogram (EEG) result from gradual structural and functional modifications that arise during the last trimester of pregnancy. Here, we review the structural changes over time that underlie the establishment of functional immature neural networks, the impact of certain anatomical specificities (fontanelles, connectivity, etc.) on the EEG, limitations in EEG interpretation, and the utility of high-resolution EEG (HR-EEG) in premature newborns (a promising technique with a high degree of spatiotemporal resolution). In particular, we classify EEG features according to whether they are manifestations of endogenous generators (i.e. theta activities that coalesce with a slow wave or delta brushes) or come from a broader network. Furthermore, we review publications on EEG in premature animals because the data provide a better understanding of what is happening in premature newborns. We then discuss the results and limitations of functional connectivity analyses in premature newborns. Lastly, we report on the magnetoelectroencephalographic studies of brain activity in the fetus. A better understanding of complex interactions at various structural and functional levels during normal neurodevelopment (as assessed using electroencephalography as a benchmark method) might lead to better clinical care and monitoring for premature neonates.
A standardised assessment scheme for conventional EEG in preterm infants
2020, Clinical Neurophysiology
Citation Excerpt :
In another study, the authors reported a substantial interobserver agreement when two experts in neonatal EEG reviewed 2-channel continuous EEG with aEEG for seizure detection in term infants, compared to a fair degree of agreement when only using the aEEG (Shah et al., 2008). However, little is known about interobserver agreement in neonatal EEG interpretation in general, particularly for background activity evaluation and for its assessment in preterm infants, for whom only specific features have been studied (Murphy et al., 2015; O'Toole et al., 2017). Murphy et al. calculated interobserver agreement for burst/interburst detection and showed that moderate levels of agreement (median Kappa from 0.53 to 0.66) were achieved among 3 observers with annotations summarizing all channels (Murphy et al., 2015).
To develop a standardised scheme for assessing normal and abnormal electroencephalography (EEG) features of preterm infants. To assess the interobserver agreement of this assessment scheme.
We created a standardised EEG assessment scheme for 6 different post-menstrual age (PMA) groups using 4 EEG categories. Two experts, not involved in the development of the scheme, evaluated this on 24 infants <32 weeks gestational age (GA) using random 2 hour EEG epochs. Where disagreements were found, the features were checked and modified. Finally, the two experts independently evaluated 2 hour EEG epochs from an additional 12 infants <37 weeks GA. The percentage of agreement was calculated as the ratio of agreements to the sum of agreements plus disagreements.
Good agreement in all patients and EEG feature category was obtained, with a median agreement between 80% and 100% over the 4 EEG assessment categories. No difference was found in agreement rates between the normal and abnormal features (p = 0.959).
We developed a standard EEG assessment scheme for preterm infants that shows good interobserver agreement.
This will provide information to Neonatal Intensive Care Unit (NICU) staff about brain activity and maturation. We hope this will prove useful for many centres seeking to use neuromonitoring during critical care for preterm infants.
Automated EEG analysis for neonatal intensive care
2019, Encyclopedia of Biomedical Engineering
The neonatal EEG is a powerful tool for assessing neurological function in critically ill neonates. It is however difficult to interpret and the expertise to translate the EEG is typically not available to the attending clinician on demand. Automated algorithms can fill the role of EEG translator and help support decision making in the neonatal intensive care unit. These algorithms can detect seizures, bursting activity and sleep state, determine the ‘normality’ of the EEG, and estimate the developmental maturity of the brain. They use a variety of EEG features combined using a range of classifiers to form a decision statistic. Several of these algorithms are approaching the benchmark of the visual interpretation of the human expert, while others are at the prototype stage. The complete development of these algorithms requires the collective efforts of engineers and clinicians and has the potential to improve clinical care and subsequent health outcomes of critically ill neonates.
It's All About the Brain—Neuromonitoring During Newborn Transition
2018, Seminars in Pediatric Neurology
Citation Excerpt :
Indeed the first ever clinical investigation of an algorithm for neonatal seizure detection, developed using machine learning, has recently completed recruitment in a multicentre study across Europe (The ANSeR study—NCT02431780). Algorithms for automated baseline neonatal EEG interpretation in term and preterm neonates are advancing at a rapid pace and it is likely that a version ready for clinical deployment will soon be available.90-96 It is clear that EEG provides objective information about newborn brain function and technological barriers that previously made it difficult to deploy in the minutes and hours after birth are no longer a concern.

View all citing articles on Scopus

View full text

Automated analysis of multi-channel EEG in preterm infants

Highlights

Abstract

Objective

Methods

Results

Conclusions

Significance

Introduction

Section snippets

Material

Burst detection

Burst detection

Acknowledgements

Electroencephalogr Clin Neurophysiol

Electroencephalogr Clin Neurophysiol

Clin Neurophysiol

Electroencephalogr Clin Neurophysiol

Phys D Nonlinear Phenomena

Pediatric Neurol

Clin Neurophysiol

Electroencephalogr Clin Neurophysiol

Clin Neurophysiol

Clin Neurophysiol

Clin Neurophysiol

Brain Dev