Statistical approaches to improving functional connectivity estimates with an application to autism

点击翻译按钮获取中文摘要

ABSTRACT Functional magnetic resonance imaging (fMRI) is used to estimate the correlations between brain regions. De- spite the many insights into brain function provided by fMRI, the field is currently experiencing a reproducibility crisis. For instance, autism spectrum disorder (ASD) is thought to be the result of disordered brain connections, but the ASD literature contains conflicting reports of both hypo- and hyper-connectivity. Participant head motion is a crucial factor, and the exclusion of high-motion participants can reduce the impacts of motion. However, motion quality control removes more than 50% of data in pediatric fMRI studies. In ASD, the most severely im- paired children tend to be removed – we are characterizing a limited part of the autism spectrum. In addition, methodological development and empirical studies have focused on functional connectivity measured at rest. Children move less during some tasks, such as watching a movie. Dynamic connectivity during a task may offer crucial insights into diseases of brain connectivity with larger effect sizes, but popular methods do not utilize task information during model fitting. The overall objective of this grant is to develop statistical methods that decrease bias and improve efficiency in functional connectivity studies. Our proposal is motivated by a resting-state fMRI study on ASD with hundreds of children from the Kennedy Krieger Institute and Johns Hopkins University. We will assess external validity in an independent test dataset that will be collected at the Marcus Autism Center and Emory University, where we will also conduct a study of engagement during a movie with social interactions. To achieve our objective, we propose the following: 1) Develop a missing data method for deconfounding to reveal functional connectivity signatures of ASD. This aim addresses sampling biases due motion quality control. 2) Develop a causal mediation framework for signal decomposition of functional connectivity and motion. This aim will be used to disentangle neural versus motion contributions to function connectivity in ASD. 3) Develop a novel covariance regression model for dynamic functional connectivity during a task. This aim will test the scientific hypothesis that the neural underpinnings of engagement differ in children with ASD compared to typically devel- oping children. We will develop software so that the proposed methods can be broadly applied to neuroimaging studies including neurological disorders and mental health. Completing these aims will 1) provide tools that will improve reproducibility in functional connectivity studies and 2) reveal the neural underpinnings of ASD. This can aid in the development of treatments and educational strategies.