Multi-Scale Graph Representation Learning for Autism Identification With Functional MRI.

Ying Chu,Guangyu Wang,Mingxia Liu,Lishan Qiao,Liang Cao

doi:10.3389/fninf.2021.802305

Abstract

Resting-state functional MRI (rs-fMRI) has been widely used for the early diagnosis of autism spectrum disorder (ASD). With rs-fMRI, the functional connectivity networks (FCNs) are usually constructed for representing each subject, with each element representing the pairwise relationship between brain region-of-interests (ROIs). Previous studies often first extract handcrafted network features (such as node degree and clustering coefficient) from FCNs and then construct a prediction model for ASD diagnosis, which largely requires expert knowledge. Graph convolutional networks (GCNs) have recently been employed to jointly perform FCNs feature extraction and ASD identification in a data-driven manner. However, existing studies tend to focus on the single-scale topology of FCNs by using one single atlas for ROI partition, thus ignoring potential complementary topology information of FCNs at different spatial scales. In this paper, we develop a multi-scale graph representation learning (MGRL) framework for rs-fMRI based ASD diagnosis. The MGRL consists of three major components: (1) multi-scale FCNs construction using multiple brain atlases for ROI partition, (2) FCNs representation learning via multi-scale GCNs, and (3) multi-scale feature fusion and classification for ASD diagnosis. The proposed MGRL is evaluated on 184 subjects from the public Autism Brain Imaging Data Exchange (ABIDE) database with rs-fMRI scans. Experimental results suggest the efficacy of our MGRL in FCN feature extraction and ASD identification, compared with several state-of-the-art methods.

Highlights

Autism spectrum disorder (ASD) is a developmental disorder that can cause major social, communication, and behavioral challenges (Simonoff et al, 2008)
We compare the proposed multi-scale graph representation learning (MGRL) approaches with three conventional networks representation methods and two Graph convolutional networks (GCNs) based methods: (1) multi-scale feature fusion based on degree centrality (DCF), (2) multi-scale feature fusion based on local clustering coefficients (LCCF), (3) multi-scale feature fusion based on closeness centrality (CCF), (4) GCN with Anatomical Labeling (AAL) atlas (GCNA), and (5) GCN with CC200 atlas (GCNC)
Compared with two single-scale GCN methods (i.e., GCNA and GCNC), the MGRL improved the classification performance by at least 3% in terms of accuracy, F1-score, recall, precision, and area under Receiver Operating Characteristics (ROC) curve (AUC). These results suggest that using multi-scale brain atlases helps boost the classification performance, when compared with that using a single atlas

Summary

Introduction

Autism spectrum disorder (ASD) is a developmental disorder that can cause major social, communication, and behavioral challenges (Simonoff et al, 2008). In 2014, the overall prevalence of autism was estimated at 16.8 per 1, 000 8-year-old children, and the prevalence of ASD reached nearly 3% in some communities (Baio et al, 2018). The current diagnosis of autism is highly dependent on traditional behavioral symptoms, which are usually subjective and can lead to neglect early symptoms and misdiagnosis (American Psychiatric Association, 2013; Lord et al, 2018). Seeking an objective biomarker for early diagnosis and timely intervention in the treatment of autism has attracted increasing attention in the field of psychiatry and neuroscience

Methods

Results

Discussion

Conclusion