Parkinson’s disease (PD) is associated with cognitive impairment and alterations in large-scale brain networks, yet the network-level mechanisms underlying these deficits remain unclear. Focusing on the default mode network (DMN), we applied graph-theoretical and statistical learning approaches to resting-state fMRI data from healthy controls, PD patients without cognitive impairment (PD-NC), and PD patients with mild cognitive impairment (PD-MCI). Across multiple analyses, a distinct hub emerged as the most robust discriminator, exhibiting reduced connectivity in PD-MCI relative to PD-NC. Global DMN topology showed subtle alterations, with decreased efficiency and clustering coefficient in PD groups, but no widespread reorganization. These results suggest that cognitive decline in PD is driven by targeted hub vulnerability rather than global network disruption. Our findings highlight the utility of meso-scale network analyses for identifying clinically relevant biomarkers in neurodegenerative disease.