A Dual-Outcome EEG-Based Psychiatric Disorder Classification Model Leveraging Hypergraph Neural Networks

Abstract

Psychiatric disorders have become a major global public health challenge, greatly affecting quality of life and daily functioning across all age groups, and placing a serious burden on healthcare systems worldwide. In real-world clinical environments, psychiatric diagnosis is still largely based on interviews, behavioral observation, and patient self-reports, while Electroencephalography (EEG) provides a non-invasive and cost-effective approach for capturing real-time brain activity related to mental and emotional states. But traditional diagnostic methods lack objective biological support and fail to represent complex brain connectivity patterns, yet conventional machine learning techniques are limited in their ability to capture the high-order relationships present in EEG signals, however accurate classification of both main disorders and specific disorders remains a significant challenge due to the heterogeneous nature of psychiatric conditions. The objective of this work is to perform dual-target classification of psychiatric disorders by identifying both main disorders and specific disorders using EEG data through Hyper Graph Neural Networks (HGNN). The study uses existing models such as Random Forest and Decision Tree, but both show relatively low performance. In contrast, the proposed HGNN driven with Tree based Generative Addictive Model (TGAM) performs much better. HGNNTGAM is proposed as the high-performance model due to its ability to capture complex high-order EEG relationships