Alterations in dynamic connectivity in Alzheimer's disease: Network changes and improved multi-stage classification.
Alzheimer's disease (AD) is a progressive neurodegenerative disorder characterized by large-scale network disruption. While static functional connectivity (sFC) has been extensively studied, dynamic functional connectivity (dFC) and its discriminative value across the AD spectrum remain insufficiently understood. In this study, resting-state functional magnetic resonance imaging (rs-fMRI) data from 174 participants in the Alzheimer's Disease Neuroimaging Initiative, including cognitively normal (CN, n = 44), subjective memory concern (SMC, n = 24), early mild cognitive impairment (EMCI, n = 46), late MCI (LMCI, n = 30), and AD (n = 30), were analyzed to assess group differences in sFC, dFC, and graph-theoretical metrics, as well as their associations with cognition. A BrainNetCNN model was further employed to evaluate the classification performance of sFC, dFC, and their combined features. The results revealed that sFC decreased across MCI stages but increased in AD, whereas dFC variability was predominantly reduced in the pre-dementia groups and increased in AD, particularly in frontal and temporal regions. Several static graph-theoretical metrics were significantly correlated with Mini-Mental State Examination (MMSE) scores, while dFC provided complementary information. In classification tasks, dFC showed higher accuracy than sFC in binary and five-class tasks, and their integration achieved the highest accuracy (CN vs. SMC: 89.6%; five-class: 82.7%). These findings suggest that dFC may provide complementary imaging information for characterizing stage-related network alterations and differentiating diagnostic groups across the AD spectrum.