Applying a data-driven approach to quantify EEG maturational deviations in preterms with normal and abnormal neurodevelopmental outcomes

Abstract

Premature babies are subjected to environmental stresses that can affect brain maturation and cause abnormal neurodevelopmental outcome later in life. Better understanding this link is crucial to developing a clinical tool for early outcome estimation. We defined maturational trajectories between the Electroencephalography (EEG)-derived 'brain-age' and postmenstrual age (the age since the last menstrual cycle of the mother) from longitudinal recordings during the baby's stay in the Neonatal Intensive Care Unit. Data consisted of 224 recordings (65 patients) separated for normal and abnormal outcome at 9-24 months follow-up. Trajectory deviations were compared between outcome groups using the root mean squared error (RMSE) and maximum trajectory deviation (δ^max). 113 features were extracted (per sleep state) to train a data-driven model that estimates brain-age, with the most prominent features identified as potential maturational and outcome-sensitive biomarkers. RMSE and δ^max showed significant differences between outcome groups (cluster-based permutation test, p < 0.05). RMSE had a median (IQR) of 0.75 (0.60-1.35) weeks for normal outcome and 1.35 (1.15-1.55) for abnormal outcome, while δ^max had a median of 0.90 (0.70-1.70) and 1.90 (1.20-2.90) weeks, respectively. Abnormal outcome trajectories were associated with clinically defined dysmature and disorganised EEG patterns, cementing the link between early maturational trajectories and neurodevelopmental outcome.

Figure 1

Pre-processing procedure for extracting a representative set of features for each recording from QS, non-QS and full-cycle EEG periods. Dimensions of the resulting output is provided below each stage, where applicable.

Figure 2

Assessing stability of the RF model and brain-age correlation with PMA over the full range of N^f. Models are tested and trained on $\mathcal{D}1$ using Leave-One-Subject-Out cross-validation. (a) Results for the trajectory $RMSEs$. Black lines denote the median values and shaded regions denote the interquartile ranges. (b) Results for $r$. (c) Scatter plot of estimated brain-age against PMA at N^f = 226. Grey line denotes estimated age = PMA and black line denotes the regression line (used to calculate $r$).

Figure 3

Overall brain-age trajectory performance on $\mathcal{D}2$ across N^f. Results are separated for normal and abnormal neurodevelopmental outcome. Lines denote the medians and shaded regions denote the interquartile ranges. Below the plot are the results for both the uncorrected test for statistically significant differences between outcome groups (using the Wilcoxon rank-sum test), as well as the cluster-based non-parametric permutation test (p < 0.05) which corrects for multiple comparisons. (a) Results for the trajectory summary metric, $RMSE$. (b) Corresponding results for δ^max. (c) Corresponding results for r. In this case, single values were calculated for each N^f so no statistical comparison was performed.

Figure 4

Illustration of the trajectory plots of $\mathcal{D}1$ for N^f = 226, separated by outcome group. The more deviant trajectories are highlighted by fitting a 95% prediction interval threshold to the normal-outcome data (left plot). The 95% prediction interval is shaded in each plot, with the trajectories highlighted in red when at least one recording exceeded this threshold. The anonymous patient ID is also provided for these cases (in blue). Grey dotted lines show the line for brain-age = PMA.

Figure 5

The top six features predominantly selected by the RF brain-age prediction models. Features are ranked according to the values of the $\mathrm{\Delta }SSE$ at N^f = 226. $\mathrm{\Delta }SSE$ is plotted for each feature across the full N^f range and shaded regions denote regions of N^f where the rank matched the rank at N^f = 226. The p-values for the PMA and outcome is also shown after a univariate regression of the individual feature data using linear mixed-effects models. ‘+ve’ and ‘−ve’ denote if the feature trends increased or decreased with PMA, respectively.

Figure 6

20 second non-QS EEG epochs extracted from recordings that formed a patient’s brain-age trajectory. On each EEG panel from top to bottom: First four EEG channels reflect the right hemisphere from anterior to posterior. Next four channels reflect left hemisphere from anterior to posterior. Channels 9–10 are right midline and channels 11–12 are left midline. The PMA and GA of the baby’s recordings (in weeks) and the overall clinical assessment of the EEG morphology (normal, dysmature or disorganised) are also given. (a) EEGs from a patient born at 30 4/7 weeks GA with a normal estimated brain-age trajectory and normal (clinically defined) EEG characteristics. (b) A patient born at 25 4/7 weeks GA with a delayed estimated brain-age trajectory and abnormal, dysmature EEG. (c) A patient born at 25 5/7. The first recording at 27 6/7 weeks PMA had an accelerated estimated brain-age trajectory and abnormal, disorganised EEG. The second recording at 30 4/7 weeks PMA had both disorganized and dysmature behaviour.