Newborn EEG connectivity analysis using time-frequency signal processing techniques

Abstract

Scalp electroencephalography (EEG) monitoring in early life is providing vital assistance to clinical management during diagnosis and treatment of developing brain disorders. It is also a useful tool for predicting long-term neurological outcomes of perinatal injuries. Continuous evolution of cerebral structures during early brain maturation imposes salient changes in the temporal morphology of newborn EEG patterns. Therefore, dealing with neonatal EEG signals is technically very challenging. The current dominant approach to newborn EEG assessment is based on visual inspection of EEG recordings by an expert. This process is highly subjective and also, very time-consuming. More importantly, it requires a high level of expertise for appropriate interpretation. On the other hand, neonatal intensive care units (NICUs) in most countries do not have ready access to high-level EEG expertise. The development of automatic newborn EEG analysis systems is therefore vital to improve international newborn health. This dissertation is focused on developing scalp EEG connectivity analysis methods for objective monitoring of newborn EEG signals. Three important characteristics of the signals are taken into consideration for the proposed methodologies: time-varying behavior, directional relationships between channels and frequency-specific fluctuations of amplitudes. It leads to a more comprehensive insight into the scalp-level electrical interactions between different cortical areas of the newborn brain in time and/or frequency domains. Two types of newborn brain abnormalities including seizures and intra-ventricular hemorrhage (IVH) and their impact on scalp EEG connectivity are investigated. Also, neonatal electric resting state networks (eRSNs) are characterized using EEG recordings of healthy fullterms, healthy preterms and preemies with IVH as well as a set of newborn functional magnetic resonance imaging (fMRI) datasets. The first contribution of this research is a new time-frequency based approach for estimating the generalized phase synchrony (GePS) among multichannel newborn EEG signals using the linear relationships between their instantaneous frequency laws. Since the underlying signals are usually multicomponent, a decomposition method like multi-channel empirical mode decomposition is used to simultaneously decompose the multi-channel signals into their intrinsic mode functions (IMFs). The results confirm that the GePS value within EEG channels increases significantly during ictal periods. A statistically consistent phase coupling is also observed within the non-seizure segments supporting the concept of constant inter-hemispheric connectivity in the newborn brain during inter-ictal periods.The second contribution is a time–frequency method for measuring directional interactions over time and frequency from scalp-recorded newborn EEG signals in a way that is less affected by volume conduction and amplitude scaling. The time-varying generalized partial directed coherence method is modified, by orthogonalization of the strictly causal multivariate autoregressive model coefficients, to minimize the effect of mutual sources. The novel measure, generalized orthogonalized PDC (gOPDC), is tested first using two simulated models with feature dimensions relevant to EEG activities. The method is then used for assessing event-related directional information flow from flash-evoked responses in neonatal EEG. For testing statistical significance of the findings, a thresholding procedure driven by baseline periods in the same EEG activity is followed. The results suggest that the gOPDC is able to remove common components akin to volume conduction effect in the scalp EEG, [1] handles the potential challenge with different amplitude scaling within multichannel signals, and [1] can detect directed information flow within a sub-second time scale in nonstationary multichannel EEG datasets. The third contribution is a novel RSN analysis framework for studying the presence of long-range spatial correlations in spontaneous brain activity of newborn cortex using scalp EEG recordings. The spatial correlations of EEG signal are found to follow a robust bimodality: during high amplitudes (high mode), the brain exhibits strong widespread correlations that disappear during low amplitudes (low mode). Moreover, a clear spatial structure with frontal and parieto-occipital sub-networks appears only towards term age. No temporal bimodality is observed in the fMRI recordings under the proposed analysis paradigm, suggesting that early EEG activity and fMRI signal reflect different mechanisms of spatial coordination. The results suggest that the early developing human brain exhibits intermittent long-range spatial connections that likely provide the endogenous guidance for early activity-dependent growth of brain networks. In summary, the techniques proposed in this dissertation contribute to the field of digital signal processing with applications to newborn EEG connectivity analysis and computer-assisted neonatal brain abnormality assessment.