Identifying Autism Spectrum Disorder Based on Individual-Aware Down-Sampling and Multi-Modal Learning

Abstract

Autism Spectrum Disorder(ASD) is a set of neurodevelopmental conditions that affect patients' social abilities. In recent years, many studies have employed deep learning to diagnose this brain dysfunction through functional MRI (fMRI). However, existing approaches solely focused on the abnormal brain functional connections but ignored the impact of regional activities. Due to this biased prior knowledge, previous diagnosis models suffered from inter-site heterogeneity and inter-individual phenotypic differences. To address this issue, we propose a novel feature extraction method for fMRI that can learn a personalized lower-resolution representation of the entire brain networking regarding both the functional connections and regional activities. Specifically, we abstract the brain imaging as a graph structure and straightforwardly downsample it to sparse substructures by hierarchical graph pooling. The down-scaled feature vectors are embedded into a population graph where the hidden inter-subject heterogeneity and homogeneity are explicitly expressed as inter-and intra-community connectivity differences. Subsequently, we fuse the imaging and non-imaging information by graph convolutional networks (GCN), which recalibrates features to node embeddings under phenotypic statistics. By these means, our framework can extract features directly and efficiently from the entire fMRI and be aware of implicit inter-individual variance. We have evaluated our framework on the ABIDE-I dataset with 10-fold cross-validation. The present model has achieved a mean classification accuracy of 85.95% and a mean AUC of 0.92, better than the state-of-the-art methods. Our code and support documents are available at github.com/jhonP-Li/ASD_GP_GCN

Identification of autism spectrum disorder based on MEWISPool and multi-modal learning

“…Another method is based on Edge-Variation graph Convolutional Networks (EV-GCNs), which can automatically integrate imaging and non-imaging data [24][25] , but only uses functional connections as the node feature. A method of GCN based on graph pooling, called GP-GCN 22 was also compared to evaluate the effect of our proposed pooling method. GP-GCN also considers functional connections and regional activities, but the pooling is based on node selection that may lose key information about the graph structure.…”

“…In this paper, we represent brain imaging as a graph according to Pan et al 22 . One hundred and ten nodes are defined based on the 110 brain regions defined by the HO atlas.…”

“…Recently, an increasing number of studies have used GCN for multi-modal fusion. Here, we construct a population graph where the connection between two subjects is determined by the similarity of their inter-phenotypic information according to Pan et al 22 . The extracted high-order features are assigned to each node.…”