Temporal variability of brain–behavior relationships in fine-scale dynamics of edge time series

Most work on functional connectivity (FC) in neuroimaging data prefers longer scan sessions or greater subject count to improve reliability of brain–behavior relationships or predictive models. Here, we investigate whether systematically isolating moments in time can improve brain–behavior relationships and outperform full scan data. We assess how behavioral relationships vary over time points that are less visible in full FC based on co-fluctuation amplitude. Additionally, we perform optimizations using a temporal filtering strategy to identify time points that improve brain–behavior relationships. Analyses were performed on resting-state fMRI data of 352 healthy subjects from the Human Connectome Project and across 58 different behavioral measures. Templates were created to select time points with similar patterns of brain activity and optimized for each behavior to maximize brain–behavior relationships from reconstructed functional networks. With 10% of scan data, optimized templates of select behavioral measures achieved greater strength of brain–behavior correlations and greater transfer of behavioral associations between groups of subjects than full FC across multiple cross-validation splits of the dataset. Therefore, selectively filtering time points may allow for development of more targeted FC analyses and increased understanding of how specific moments in time contribute to behavioral prediction.