Motifs-Based Multi-Scale Graph Convolution Learning Framework for Attention Deficit/Hyperactivity Disorder Classification

Attention Deficit/Hyperactivity Disorder (ADHD) is a highly prevalent neurodevelopmental disorder. Recent studies suggest functional brain networks could provide crucial insights for diagnosing ADHD. However, most of these studies primarily focus on functional connections, often overlooking complex higher-order spatial topological relationships among brain regions. To address this limitation, a Motifs-based Multi-scale Graph Convolution (MMGC) model is proposed to enhance feature representations by aggregating spatial features from neighboring nodes across various higher-order relationships. Our experiments indicate the effectiveness of the proposed network, achieving an average accuracy of $\mathbf{7 2. 0 4} \%$, thereby outperforming traditional brain network classification methods.