Associations Between Attention Performance and Graph Signal Processing Features of Multi-State EEG

Attention is a fundamental component of cognition, and its variability is closely linked to learning efficiency, executive functioning, and a wide range of neuropsychiatric conditions. EEG is widely used to investigate neural dynamics associated with attention. At the same time, Graph Signal Processing (GSP) provides a framework for graph-spectral signal analysis that embeds the brain’s connectivity topology into the computation, thereby enabling the extraction of graph-domain features that traditional time–frequency methods cannot adequately characterise. Although GSP has shown potential in describing network-level dynamics from resting-state EEG, resting data alone may not fully reveal the task-related neural mechanisms that directly support attentional processing. The symbol search task is susceptible to individual differences in selective attention and processing speed, and the behavioural distinction between correct and error trials offers valuable information for examining network changes across different attention states. However, global dynamical brain network features from multi-state EEG have not been sufficiently explored in the context of attention research. In this study, we analysed EEG data collected during the resting state, together with EEG recorded during correct and error-symbol search trials, enabling a direct comparison between intrinsic neural activity under varying attentional demands. We extracted a comprehensive set of GSP-based graph spectral features across these multi-state conditions. We systematically compared feature groups across states, assessing both statistical differences and their ability to discriminate between high- and lowattention participants using classification analyses. We further examined their associations with attentional performance to provide a more comprehensive characterisation of the graph spectral features that reflect individual differences in attention.