Multi-View Guided Hierarchical Graph Neural Network for ASD Detection

In recent years, graph neural networks (GNN) based on functional magnetic resonance imaging (fMRI) have shown great potential in the study of Autism Spectrum Disorder (ASD). To address the limitations of existing deep learning methods in effectively capturing short-term brain dynamics, this paper proposes a Multi-View Spatiotemporal Feature-Guided Hierarchical Graph Neural Network. The model first employs a KL-Divergence-based mechanism to select key temporal features, followed by an alignment module to align features across multiple views. Additionally, an interactive graph convolutional layer is designed to fuse imaging and non-imaging information, aiming to enhance classification accuracy. Experimental results on the Autism Brain Imaging Data Exchange (ABIDE) dataset demonstrate that our model achieves an average accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 1. 6 2 \%}$</tex>, confirming its effectiveness in accurate ASD diagnosis.