Employing Temporal Properties of Brain Activity for Classifying Autism Using Machine Learning

Abstract

Exploration of brain imaging data with machine learning methods has been beneficial in identifying and probing the impacts of neurological disorders. Psychopathological ailments that disrupt brain activity can be discerned with the help of resting-state functional magnetic resonance imaging (rs-fMRI). Research has revealed that brain connectivity is dynamic in nature and that its dynamic properties are affected by brain disorders. In the literature, numerous approaches have been proposed for identifying the presence of Autism Spectral Disorder (ASD), yet most of them do not consider brain dynamics in their diagnostic process. Significant amount of knowledge can be procured by taking the evolution of brain connectivity over time into account. In this work, we propose a new approach that leverages brain dynamics in the classification of autistic and neurotypical subjects using rs-fMRI data. We examined the proposed method on a large multi-site dataset known as ABIDE (Autism Brain Imaging Data Exchange) and have achieved state-of-the-art classification results with an accuracy of 73.6%. Our work has shown that taking the temporal properties of brain connectivity into account improves the classification performance.