Association between structural disconnection and early cognitive impairment after acute ischemic stroke: A network-based statistical study

Abstract

Post-stroke cognitive impairment (PSCI) affects up to 75% of patients acutely and strongly predicts dementia, yet the acute-phase structural network signature remains unclear. We included 589 first-ever acute ischemic stroke patients (≤ 7 days) and classified them into PSCI and non-PSCI groups using the acute-phase MoCA. Individualized structural disconnection matrices (135 nodes) were constructed from DWI lesion masks and the HCP-842 normative tractography atlas (derived from 842 healthy adults), representing indirect disconnectome estimates rather than patient-specific DTI tractography. Network-Based Statistic (NBS) identified group-different subnetworks, followed by binary graph-theoretical analysis. In exploratory full-sample analysis, a left-lateralized subnetwork of 25 nodes and 42 edges (P_FWE < 0.001) showed prefrontal-striatal radiating disconnection converging on the left caudate nucleus, and greater subnetwork disconnection strength correlated with worse acute MoCA (r = −0.27, P < 0.001). To address circularity, repeated internal discovery–validation analyses using five random 70/30 splits showed that the direction of the SubNet_Total_SDC–MoCA association was reproduced in held-out test sets, where SubNet_Total_SDC was negatively associated with MoCA (partial r = −0.265 to −0.150, test-set ΔR² = 0.014–0.056). The left caudate nucleus was reproducibly identified as the anatomical convergence point across all training samples. These findings suggest that acute PSCI is associated with a cortico-subcortical radial disconnection pattern converging on the left caudate nucleus under the present parcellation.