Global Connectivity-driven EEG Modeling: A Full-channel Temporal Sliding Framework for Alzheimer's Disease Detection

Alzheimer's disease (AD) is linked to disrupted interactions across distributed brain regions.While electroencephalography (EEG) is a promising tool for automated AD detection, many existing methods fail to capture global, whole-scalp interactions.This study proposes the Full-Channel Temporal Sliding Convolutional Network (TSCNet), a novel architecture designed to jointly capture global inter-channel dependencies and temporal dynamics.By using a convolutional kernel that spans all electrodes while sliding along the temporal axis, TSCNet learns comprehensive connectivity and oscillatory patterns.Evaluated on harmonized EEG datasets represented in a standardized 19-channel montage, TSCNet achieved an average accuracy of 96.95%, sensitivity of 98.57%, and specificity of 95.00%, significantly outperforming standard convolutional baselines.These results demonstrate that explicitly modeling whole-scalp connectivity provides a robust spatial representation for AD detection.This framework establishes a high-performing baseline for the further refinement of heterogeneous temporal dynamics in neurodegenerative research.