Resting-state functional connectivity (RSFC) and networks (RSNs) provide insight into large-scale brain organization and its disruption in neurological disease. RSNs are most commonly assessed using fMRI, yet its translational use is constrained by high cost, motion sensitivity, and limited feasibility for repeated measurements. Functional near-infrared spectroscopy (fNIRS) offers a portable alternative, but its reliability for RSFC and RSN mapping remains insufficiently established. Near whole-head fNIRS data and fMRI-BOLD signals of corresponding cortical regions were extracted, based on which RSN organization was compared across two independent cohorts of 31 participants each. Cross-modal convergence and divergence were assessed using bivariate and partial correlations across multiple network levels. Edgewise analyses revealed substantial modality differences with bivariate correlations (50-61% of edges), which were markedly reduced using partial correlations (<3%). Group-averaged connectivity patterns showed moderate cross-modal similarity (r ≈ 0.37). At nodal level, net strength, local efficiency, and path-length differed substantially between modalities, while normalized strength and assortativity were largely comparable. Across nodes, graph-metric distributions derived from group-averaged matrices were broadly similar for normalized strength, assortativity, local efficiency, and path length (rho ≈ 0.27-0.5). At network-level, fNIRS-derived modules significantly overlapped with fMRI modules, particularly based on bivariate correlations, identifying default mode, attentional, executive, salience, sensorimotor, and visual networks (Jaccard ≈ 0.27-0.5). Overall, fNIRS captured key features of large-scale RSFC and RSN organization observed with fMRI, supporting meaningful cross-modal correspondence and translational utility. While partial correlations enhanced edge-level agreement, they attenuated nodal and modular recovery, suggesting greater suitability for targeted connectivity analyses rather than whole-network characterization.
Kotsogiannis et al. (Tue,) studied this question.