Resting-State and Task Functional Magnetic Resonance Imaging Network Topology Metrics With no Threshold Selection to Predict Cognition.

Network topology measures characterise brain networks' organisation. Graph theoretical approaches have shown fMRI topology metrics' association with cognitive performance. Because arbitrary connectivity threshold selection biases such metrics, alternatives including the minimum spanning tree (MST) and novel measures following principles of persistent homology were proposed. The present study compared alternative and graph theoretical metrics in association with cognition for resting-state and task-fMRI. Functional connectivity matrices were computed from Human Connectome Project (Young Adult) fMRI scans during resting-state, working memory (WM), gambling, language, motor, relational processing, social cognition, and movie-watching conditions. Global efficiency, clustering coefficient (at three thresholds), diameter, leaf fraction (LF), backbone strength (BS), and cycle strength were measured. Each was tested in association with cognitive test scores. ResultsBS significantly predicted general cognitive performance, specifically progressive matrices score, composite fluid and crystallised cognition, vocabulary, spatial orientation, and WM. Diameter significantly predicted WM. WM task BS outperformed the predictive performance of graph theory measures, but not at rest, where MST LF outperformed other measures. Stronger associations were observed between cognitive test scores and topology measures derived from task-based fMRI, especially the N-Back task, as opposed to resting-state fMRI. Among task-based topology measures, BS was the most strongly related to cognition.