EEG-Based Detection of Autism Spectrum Disorder Using Hypergraph Convolutional Networks with Explainable AI Interpretability

Abstract

The occurrence of autism spectrum disorder (ASD) has been increasing in recent years, emphasizing the importance of Early diagnosis and intervention. Early detection provides a promising solution for ASD affected people to lead their life individually. Existing methods for ASD diagnosis face challenges in time complexity in traditional manual observation while contemporary models based on machine learning are sometimes viewed with skepticism for hidden patterns. Several studies suggest that EEG-based diagnostic methods are superior to traditional techniques since EEG uncovers neural dynamics associated with ASD. To address these issues, this work utilizes a Hybrid Graph Convolutional Network (HGCN) approach for EEG signal analysis to capture the intricate details of connectivity topology with a view of enhancing the classification performance. To counter the issue of trust and transparency, Explanatory AI (XAI) is integrated into the model to assist clinicians and researchers to understand the working of the model and its predictions. The integration of HGCN and XAI is highly advantageous since it offers a solution where the model has not only high accuracy, but also easy interpretability, making it applicable to clinical diagnosis.