EEG-based Brain Connectivity Analysis for Identifying Neurodevelopmental Disorders

Abstract

This dissertation aims to identify the neurological biomarkers that could assist in providing reliable, automated and objective prediction of neurodevelopmental disorders (NDDs) in early infancy. Quantitative electroencephalography analysis (qEEG), mainly phase synchronisation-based functional brain connectivity estimated using phase locking value (PLV) and weighted phase lag index (WPLI), were investigated to deduce whether it can be used for the early prediction of such disorders. The resulting connectivity network was quantitatively characterised using complex graph-theoretical features, namely transitivity, global efficiency, radius, diameter, and characteristic path length. These features were then fed into the machine learning algorithms such as linear discriminant analysis (LDA), support vector machine (SVM), decision tree and k-nearest neighbour to examine their discriminant capability in classifying /predicting NDDs. The proposed framework has gained initial validation in classifying autism spectrum disorders (ASD) from an experimentally obtained EEG data set of 24 children. Then, the framework was utilised to predict the appearance of cerebral palsy (CP) at two years of age. The EEG data were recorded within the first week after birth from a cohort of infants born with hypoxic-ischaemic encephalopathy (HIE). The exploration results revealed that the proposed analytical methodology successfully predicted the infants that would develop CP with a performance of 84.6% accuracy, 83% sensitivity, 85% specificity, 84% balanced accuracy and 0.85 area under the curve (AUC) in the delta band, with a close result also obtained in the theta and alpha bands. The WPLI and graph parameters were then used to predict the cognitive scores of infants born with HIE by developing the regression framework correlating these EEG features and a cognitive profile completed in a follow-up assessment at two years of age. The regression analysis showed that the radius feature yielded the best performance (root mean square error (RMSE)= 16.78, mean absolute error (MAE)= 12.07 and R-squared= 0.24). Although this study has successfully demonstrated that the qEEG features could be considered potential biomarkers for identifying the brain deficits causing the NDDs, it has a certain limitation due to the size of the data set. It needs to be validated on large trials with a statistically significant population.