Functional brain networks involved in the Raven's standard progressive matrices task and their relation to theories of fluid intelligence

A dimensionality reduction method was used to determine the task-timing-related functional brain networks underlying the Raven's Standard Progressive Matrices (RSPM), a non-verbal estimate of fluid intelligence (Gf). We identified five macro-scale task-based blood-oxygen-level-dependent (BOLD)-signal brain networks and interpreted their network-level task-induced BOLD changes to provide functional interpretations separately for each network. This led to new observations about the brain networks underlying the RSPM: (1) the multiple demand network (MDN) for solution searching peaked early in the trial (-9 s peak), followed by response (RESP) for response selection (-12 s), and re-evaluation (RE-EV) for solution checking (-18 s peak), (2) high activity in the MDN was correlated with high activity in the later-peaking RE-EV network, proposed to underpin cooperative solution searching (MDN) and checking (RE-EV) processes, and (3) high activity in the MDN in all conditions was associated with low accuracy in the hard RSPM condition, suggesting that those with lower performance on hard problems allocate more resources into solution-searching across all conditions. These findings corroborate the MDN's significance in Gf solution searching, and add the RE-EV network as playing an important checking role, providing overlap with the proposed abstraction/elaboration and hypothesis testing phases of the Parieto-Frontal Integration Theory (P-FIT). Therefore, this set of results not only supports past theoretical work on the brain networks underlying Gf and the RSPM task, but extends it by providing more complete anatomical, temporal, and functional information based on a set of brain task-based networks which replicate over many tasks.