Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.770898
Title: Applications of multi-way analysis for characterizing paediatric electroencephalogram (EEG) recordings
Author: Kinney-Lang, Eli W.
ISNI:       0000 0004 7655 0797
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
This doctoral thesis outlines advances in multi-way analysis for characterizing electroencephalogram (EEG) recordings from a paediatric population, with the aim to describe new links between EEG data and changes in the brain. This entails establishing the validity of multi-way analysis as a framework for identifying developmental information at the individual and collective level. Multi-way analysis broadens matrix analysis to a multi-linear algebraic architecture to identify latent structural relationships in naturally occurring higher order (n-way) data, like EEG. We use the canonical polyadic decomposition (CPD) as a multi-way model to efficiently express the complex structures present in paediatric EEG recordings as unique combinations of low-rank matrices, offering new insights into child development. This multi-way CPD framework is explored for both typically developing (TD) children and children with potential developmental delays (DD), e.g. children who suffer from epilepsy or paediatric stroke. Resting-state EEG (rEEG) data serves as an intuitive starting point in analyzing paediatric EEG via multi-way analysis. Here, the CPD model probes the underlying relationships between the spatial, spectral and subject modes of several rEEG datasets. We demonstrate the CPD can reveal distinct population-level features in rEEG that reflect unique developmental traits in varying child populations. These development-affiliated profiles are evaluated with respect to capturing structures well-established in childhood EEG. The identified features are also interrogated for their predictive abilities in anticipating new subjects' ages. Assessing simulations and real rEEG datasets of TD and DD children establishes the multi-way analysis framework as well suited for identifying developmental profiles from paediatric rEEG. We extend the multi-way analysis scheme to more complex EEG scenarios common in EEG rehabilitation technology, like brain-computer interfaces. We explore the feasibility of multi-way modelling for interventions where developmental changes often pose as barriers. The multi-way CPD model is expanded to include four modes- task, spatial, spectral and subject data, with non-negativity and orthogonality constraints imposed. We analyze a visual attention task that elucidates a steady-state visual evoked potential and present the advantages gained from the extended CPD model. Through direct multi-linear projection, we demonstrate that linear profiles of the CPD can be capitalized upon for rapid task classification sans individual subject classifier calibration. Incorporating concepts from the multi-way analysis scheme with child development measured by psychometric tests, we propose the Joint EEG Development Inference (JEDI) model for inferring development from paediatric EEG. We utilize a common EEG task (button-press) to establish a 4-way CPD model of paediatric EEG data. Structured data fusion of the CPD model and cognitive scores from psychometric evaluations then permits joint decomposition of the two datasets to identify common features associated with each representation of development. Use of grid search optimization and a fully cross-validated design supports the JEDI model as another technique for rapidly discerning the developmental status of a child via EEG. We then briefly turn our attention to associating child development as measured by psychometric tests to markers in the EEG using graph network properties. Using graph networks, we show how the functional connectivity can inform on potential developmental delays in very young epileptic children using routine, clinical rEEG measures. This establishes a potential tool complementary to the JEDI model for identifying and inferring links between the established psychometric evaluation of developing children and functional analysis of the EEG. Multi-way analysis of paediatric EEG data offers a new approach for handling the developmental status and profiles of children. The CPD model offers flexibility in terms of identifying development-related features, and can be integrated into EEG tasks common in rehabilitation paradigms. We aim for the multi-way framework and associated techniques pursued in this thesis to be integrated and adopted as a useful tool clinicians can use for characterizing paediatric development.
Supervisor: Escudero Rodriguez, Javier ; Polydorides, Nicholas ; Auyeung, Bonnie Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.770898  DOI: Not available
Keywords: electroencephalogram recordings ; EEG data ; canonical polyadic decomposition ; CPD ; paediatric EEG recordings ; CPD framework ; rEEG ; development-affiliated profiles ; JEDI model ; graph networks ; CPD model
Share: