Complexity and synchronisation analysis of electroencephalogram signals for early prediction of neurodevelopmental disorders

Abstract

Neurodevelopmental disorders (NDDs) are a group of neurological disorders emerging during early development and impacting higher-level brain function. Early identification of NDDs raises the possibility of improving outcomes. However, NDDs are still clinically detected through a subjective evaluation approach that lacks biological evidence and requires an extended period to identify the disorder. This thesis aims to identify biomarkers from the electroencephalogram (EEG) for abnormal brain dynamics and utilise these biomarkers for NDDs prediction. Nonlinear time series analysis methods, mainly complexity (entropies) and synchronisation (node degree of the phase-lag index) features are investigated for this purpose due to their ability to reflect brain dynamics. A machine learning framework that combines advanced nonlinear EEG processing methods for NDDs identification has been developed and evaluated. The proposed framework was first used with a dataset of children with autism spectrum disorder (ASD) and controls to validate its classification efficacy. This exploration resulted in high performance, suggesting complexity features as biomarkers for ASD prediction. The framework was then used to explore at-birth EEG characteristics of infants with Hypoxic-Ischemic Encephalopathy (HIE) who later developed cerebral palsy (CP). High performance has been reached, and the proposed features were reported as potential biomarkers for early CP prediction. A regression model has also been developed to explore the correlation between the EEG characteristics of the HIE infants and their two-year cognitive scores. The results suggest the proposed features as biomarkers for early cognitive function prediction. Finally, the framework was used with a dataset of individuals with a major depressive disorder to validate its ability to predict their depression severity. Compared to the state-of-the-art research, an acceptable regression performance has been reached, and the proposed features have been suggested as biomarkers for depression severity prediction. This work lays the foundation for evidence-based decision-making applications for early prediction of CP and cognitive outcomes of infants with HIE, paving the way for establishing tailored intervention programs at an appropriate point during development to improve the outcomes.

More information

Identifiers

Catalogue record

Export record