Multi-View United Transformer Block of Graph Attention Network based Autism Spectrum Disorder Recognition

D. Darling Jemima ^1* A.Grace Selvarani ² Daphy Louis Lovenia ³

• ¹ Sri Krishna College of Technology, Coimbatore, India
• ² Sri Ramakrishna Engineering College, Coimbatore, Tamil Nadu, India
• ³ Karunya Institute of Technology and Sciences, Coimbatore, Tamil Nadu, India

Identification of Autism Spectrum Disorder (ASD) proves challenging because ASD is complex, heterogeneous, and early detection with intervention is much more effective in improving the future. Recently, deep learning has shown promise as an approach towards the improvement of diagnosing ASD through data analysis of neuroimaging. To meet the limitations within existing approaches, in this research study, a novel model has been incorporated-the Multi-View United Transformer Block of Graph Attention Network or MVUT_GAT. MVUT_GAT combines the advantages of multi-view learning and attention mechanisms to extract subtle patterns from both structural and functional MRI data. With the Autism Brain Imaging Data Exchange (ABIDE) dataset, our comprehensive analysis demonstrates that MVUT_GAT outperforms the MVS_GCN model with a +3.40% improvement in accuracy. This improvement may indicate the potential of MVUT_GAT in advancing ASD identification. Better accuracy and consistency of the model are strong contributing factors to early diagnoses. Improved early diagnosis enables timely intervention and support for ASD patients. Also, MVUT_GAT has an interpretability bridge between complexity in deep learning models and clinical application by identifying biomarkers associated with ASD. This advance in knowledge about recognition and management of ASD enhances the chances of better outcomes and quality of life for individuals affected by the disorder.