EEG affective state profiling for understanding and analyzing children`s brain development

Abstract

For the past 3 decades, the ‘Theory of Mind’ (ToM) seems to be the most established theory in explaining children with brain developmental disorder (BDD). The Theory of Mind is strongly linked to the cognitive perspective of emotion, where children with Autism Spectrum Disorder (ASD) are incapable of understanding the affective state of other people. This work focuses on understanding and analyzing children’s brain development, specifically those with ASD prior to profiling their affective states. The electroencephalogram (EEG) appears to be the best candidate for investigating human’s affective states due to its high temporal resolutions compared to fMRI and PET scans. This provides a motivation for proposing an EEG affective profiling system for the pre-screening of ASD. EEG data are collected from 30 typically developing children aged between 4-6 years during (a) resting state, (b) while watching emotionally related facial expressions and (c) during an executive function (EF) tasks. There are several challenges with the EEG affective profiling system for the purpose of pre-screening ASD identified in this thesis. Studies are heavily dependent on the qualitative or visual nature of the brain signals, rather than focusing on the quantitative features of the EEG. There is also a lack of EEG data corpus for analyzing children’s affective states.The problem is further amplifiedby the need forthe design of algorithms indetectingaffective responses, where most researchers have neglected the possibility of incorporating recent breakthroughsfrom the field of psychology. The principal contributions of this thesis are: EEG experimental stimuli and experimental protocol based on ToM for eliciting children’s affective responses, an EEG data corpus for the investigation of children’s brain developmental disorder and multi-paradigm approach that integrates different statistical and computational methodsfor children’s affective profiling. The multi-paradigm approach is divided into three parts; the reliability and sub-band analysis of the EEG data corpus, the affective recognition system using artificial neural network (ANN) and the valence-arousal mapping system based on the recalibrated Speech Affective Space Model (rSASM) and the psychological 12-Point Affective Circumplex (12-PAC). The use of 12-PAC model is a novel approach where a dimensional emotion model proposed by psychologists is adapted for the purpose of multi-label, real-valued classification. Results indicated that higher precision of affective recognition is achieved using 12-PAC compared to the rSASM. Further analysis revealed that children’s affective states are unique that points towards personalized testing. The knowledge obtained through the EEG affective profiling may greatly assist in the pre-screening of brain developmental disorder.