Identify abnormal functional connectivity of resting state networks in Autism spectrum disorder and apply to machine learning-based classification

Abstract

Autism spectrum disorder (ASD) patients are often reported altered patterns of functional connectivity (FC) on resting-state functional magnetic resonance imaging (rsfMRI) scans. However, the results in similar brain regions were inconsistent. In this study, we first investigated statistical differences in large-scale resting-state networks (RSNs) on 192 healthy controls (HCs) and 103 ASD patients by using independent component analysis (ICA). Second, an image-based meta-analysis (IBMA) was applied to discover the consistency of spatial patterns from different sites. Last, utilizing these patterns as features, we used Support Vector Machine (SVM) classifier to identify whether a subject was suffering from ASD or not. As a result, six RSNs were obtained with ICA. In each RSN, we identified altered functional connectivity between ASD and HC across the multi-site data. We calculated the area under the receiver operating characteristic curve plots (AUC) to determine the classification performance. The AUC value of classification reaches 0.988. In conclusion, the present study indicates that intrinsic connectivity patterns produced from rsfMRI data could yield a possible biomarker of ASD and contributed to the neurobiology of ASD.

Keywords: Autism spectrum disorder; Image-based meta-analysis; Independent component analysis; Resting-state functional magnetic resonance imaging; Support Vector Machine.