Altered Driving Structures and Controllability Patterns of Brain Effective Networks in Mild Cognitive Impairment and Alzheimer’s Disease

Abstract

Highlights What are the main findings? Diagnostic-group differences in threshold-derived driving structures were observed in structurally constrained directed brain effective networks, with default-mode regions acting as common drivers and additional sensorimotor drivers in MCI and frontoparietal drivers in AD. Controllability patterns varied by analytic scale: whole-brain, default-mode, and frontoparietal indices showed a nonmonotonic NC–MCI–AD pattern, whereas hub-level controllability showed the opposite pattern; regional node controllability was associated with cognitive scores. What are the implications of the main findings? These findings suggest that MCI and AD are associated with changes in the brain’s network control architecture, providing insight into altered directed information propagation during cognitive decline. The controllability patterns of multimodal brain effective networks may provide a framework for characterizing network-level differences across the AD clinical spectrum and for motivating future longitudinal validation. Abstract Background: Alzheimer’s disease (AD) and mild cognitive impairment (MCI) are associated with abnormalities in brain networks. However, differences in the control architecture of directed brain effective networks among normal controls (NC), patients with MCI, and patients with AD remain insufficiently characterized. This study investigated diagnostic-group differences in driving structures and controllability patterns in structurally constrained brain effective networks. Methods: Multimodal neuroimaging data, including diffusion MRI and resting-state functional MRI, were used to construct brain effective networks in the NC, MCI, and AD groups. Directed interregional interactions were estimated using multivariate autoregressive modeling under structural connectivity constraints. A structural controllability framework based on maximum matching was then applied to identify group-level driving nodes and driving edges. Controllability index was compared across groups separately at the whole-brain, resting-state network (RSN), and regional levels. Results: Group-level driving structures differed among the NC, MCI, and AD groups. Driving nodes were mainly distributed within the default mode network, with additional somatosensory and motor network drivers in MCI and a frontoparietal network driver in AD. The whole-brain controllability index decreased from NC to MCI and increased from MCI to AD. Similar nonmonotonic patterns were observed in the default mode and frontoparietal networks, whereas the somatosensory and motor and visual networks showed different group patterns. Regional node controllability was negatively associated with indegree and positively associated with outdegree, and driving nodes were more likely to correspond to outdegree hubs. Hub-level controllability showed a pattern opposite to that of the whole brain. Several regional indices were associated with cognitive scores, although these pooled cross-sectional associations may partly reflect diagnostic-group separation. Conclusions: MCI and AD were associated with differences in driving structures and controllability patterns in directed brain effective networks. These findings provide preliminary network-level observations on altered directed information propagation across the AD clinical spectrum and require validation in larger, independent, and longitudinal cohorts.