Determining the Functional Neural Relation Between Autism and Attention-Deficit Hyperactivity Disorder in Children through Machine Learning Analysis of Electroencephalography

Abstract

Autism and attention-deficit hyperactivity disorder (ADHD) are two neurodevelopmental conditions (NDCs) thought to arise from altered trajectories of childhood brain development. The overlap of behavioral traits and clinical co-occurrence of autism and ADHD has led to notions that the two may share similarities in associated biological profiles, including distinctive characteristics of neural function. This dissertation explores the intersection of these two NDCs from a perspective of associated neural activity through decomposition of a rich dataset of neural time series recordings. Neural signatures of autism and ADHD diagnosis are identified using a novel deep learning approach that decomposes brain activity into a compressed representation of biologically plausible, interpretable brain networks. Critically, this novel approach mitigates the influence of individual-specific signatures in representations shared across individuals, thus permitting the analysis of data collected from modestly sized cohorts such as the one highlighted here. Learned signatures reveal that autism and ADHD are characterized by non-overlapping neural features. Furthermore, we demonstrate the robustness of these learned signatures by accurately predicting their intersection. We show that the neural features associated with autism + ADHD co-occurrence can be significantly recovered through an additive, linear combination of the features associated with the two conditions. This result demonstrates that the interpolating space between autism and ADHD representations provides utility in describing the co-occurring condition. This suggests that autism and ADHD are not isolated conditions, rather, there is spectrum of neural function relating the two. The results detailed in this dissertation represent progress in the characterization of the brain activity that underly autistic and ADHD-associated behaviors and the further refinement of the neurobiological relationship between these two common NDCs.