Longitudinal Alterations in Morphometric Inverse Divergence Networks Among Diabetes Patients with Progressive Cognitive Decline.
Diabetes mellitus (DM) is associated with an elevated risk of cognitive decline, though trajectories are heterogeneous. This study investigated whether a novel, clinically applicable measure of brain network integrity, the Morphometric Inverse Divergence (MIND) network, could differentiate and predict cognitive progression in DM.
We retrospectively analyzed 101 DM participants (41 cognitively normal, 60 with mild cognitive impairment) from the Alzheimer's Disease Neuroimaging Initiative, classifying them into stable (DM_S, n=64) or decline (DM_D, n=37) group based on longitudinal diagnostic conversion. MIND networks were constructed from multiple cortical morphological features derived from T1-weighted MRI and graph theory measurements were further analyzed. Using network-based statistics (NBS) and its extension NBS-predict, we tested whether subject-level connectomes were associated with long-term DM-related cognitive worsening.
At baseline, DM_D individuals exhibited significantly lower cognitive scores and a focal subnetwork of disrupted morphometric similarity, primarily involving temporal regions. Longitudinally, DM_D individuals showed a more targeted pattern of network change that significantly altered global efficiency, local efficiency, and path length exclusively, while stable individuals, the brain underwent more widespread changes. Crucially, baseline MIND networks significantly predicted long-term cognitive progression status (accuracy = 63.1%, p = 0.034). The predictive subnetwork was rich in transmodal connections involving the temporoparietal, default mode, and limbic networks.
These findings indicate that cognitive decline in DM is preceded by specific disruptions in the brain's structural connectome. The MIND method shows promise as a network-based biomarker for identifying at-risk individuals and predicting cognitive trajectory, potentially driving advanced network analyses toward real-world applicability.
We retrospectively analyzed 101 DM participants (41 cognitively normal, 60 with mild cognitive impairment) from the Alzheimer's Disease Neuroimaging Initiative, classifying them into stable (DM_S, n=64) or decline (DM_D, n=37) group based on longitudinal diagnostic conversion. MIND networks were constructed from multiple cortical morphological features derived from T1-weighted MRI and graph theory measurements were further analyzed. Using network-based statistics (NBS) and its extension NBS-predict, we tested whether subject-level connectomes were associated with long-term DM-related cognitive worsening.
At baseline, DM_D individuals exhibited significantly lower cognitive scores and a focal subnetwork of disrupted morphometric similarity, primarily involving temporal regions. Longitudinally, DM_D individuals showed a more targeted pattern of network change that significantly altered global efficiency, local efficiency, and path length exclusively, while stable individuals, the brain underwent more widespread changes. Crucially, baseline MIND networks significantly predicted long-term cognitive progression status (accuracy = 63.1%, p = 0.034). The predictive subnetwork was rich in transmodal connections involving the temporoparietal, default mode, and limbic networks.
These findings indicate that cognitive decline in DM is preceded by specific disruptions in the brain's structural connectome. The MIND method shows promise as a network-based biomarker for identifying at-risk individuals and predicting cognitive trajectory, potentially driving advanced network analyses toward real-world applicability.