Diagnosis of Autism Spectrum Disorder Using Context-Based Pooling and Cluster-Graph Convolution Networks

Abstract

Autism spectrum disorder (ASD) is a neurological disorder that causes impairment in the healthy development of a subject’s analytic and social skills. Several studies exist in the literature on the diagnosis of ASD using machine learning, kernel-based learning, and deep learning techniques. Most of these depend on the correlation values between regions of interest in a human brain and ignore the non-anatomical phenotypical data associated with the subjects. This leads to non-uniform measurements concerning various data sources. As an attempt to bridge this gap, this paper considers both anatomical and phenotypical features. We propose a new graph-based machine learning architecture which uses graph-theoretic biomarkers for diagnosing ASD. The model uses node ranking, message passing mechanism, graph embedding, and cluster-graph convolution network for classification. Further, the model is implemented on a benchmark real-time dataset containing MRI, fMRI, and phenotypical data collected through multiple international resources. The results obtained show that the proposed model outperforms the state-of-the-art models. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.