Objective GAD Quantitative Assessment via EEG Functional Connectivity and Conv_gMLP Model

Generalized anxiety disorder (GAD) is a prevalent psychiatric disorder, yet its severity assessment relies on subjective clinical evaluations and patient self-reports, lacking objective methods. This study introduces a pioneering approach to quantitatively as-sess GAD severity by integrating resting-state EEG data with advanced artificial intel-ligence techniques, addressing the critical need for objective diagnostic tools. A total of 39 healthy controls and 80 patients with GAD were recruited, and 10-minute rest-ing-state EEG data were collected from each participant. Functional connectivity (FC) features were extracted from EEG segments across time windows of 2–10 seconds and processed using a novel deep learning framework, the Conv_gMLP model, which lev-erages a gated Multi-Layer Perceptron (gMLP) architecture for precise severity predic-tion. Our innovative Conv_gMLP model achieved a remarkable mean absolute error of 0.32 ± 0.07 within a 10-second window, significantly outperforming existing models. Notably, enhanced FC between the frontal and temporal lobes, particularly the ampli-fication of beta rhythms, emerged as a critical biomarker for GAD severity. These findings highlight the Conv_gMLP model’s superior performance and reliability as an objective tool for GAD assessment. By emphasizing frontal and temporal connectivity and beta rhythms, this study underscores the potential of FC-based feature selection to refine clinical evaluation and treatment approaches for GAD.