Macro-Micro structural integration for characterizing severity-related structural alterations in mild cognitive impairment.
Authors
Affiliations (13)
Affiliations (13)
- School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China; Queen Mary School Hainan, Beijing University of Posts and Telecommunications, Hainan, China. Electronic address: [email protected].
- School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China. Electronic address: [email protected].
- School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China. Electronic address: [email protected].
- School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China. Electronic address: [email protected].
- Department of Radiology, Qilu Hospital of Shandong University; Qilu Medical Imaging Institute of Shandong University, Jinan, China; Research Institute of Shandong University, Magnetic Field- Free Medicine & Functional Imaging, Jinan, China; Shandong Key Laboratory: Magnetic Field- Free Medicine & Functional Imaging (MF), Jinan, China.
- Department of Radiology, the Second Medical Centre, National Clinical Research Centre for Geriatric Diseases, Chinese PLA General Hospital, Beijing, China. Electronic address: [email protected].
- Department of Neurology, the Second Medical Centre, National Clinical Research Centre for Geriatric Diseases, Chinese PLA General Hospital, Beijing, China. Electronic address: [email protected].
- Department of Radiology, Xuanwu Hospital of Capital Medical University, Beijing, China.
- Department of Neurology, the Second Medical Centre, National Clinical Research Centre for Geriatric Diseases, Chinese PLA General Hospital, Beijing, China.
- Department of Neurology, Xuanwu Hospital of Capital Medical University, Beijing, China; School of Biomedical Engineering, Hainan University, Haikou, China; Center of Alzheimer's Disease, Beijing Institute for Brain Disorders, Beijing, China.
- Department of Neurology, Tianjin Huanhu Hospital, Tianjin, China.
- School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China. Electronic address: [email protected].
- School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China. Electronic address: [email protected].
Abstract
Mild cognitive impairment (MCI) is a clinically heterogeneous condition characterized by substantial variation in structural abnormalities and cognitive impairment. However, existing neuroimaging-based approaches are limited by insufficient modeling of coordinated gray-white matter alterations and vulnerability to site effects in multi-center data. Here, we propose a Macro-Micro Structural Integration (MMSI) framework for characterizing structural-deviation heterogeneity and identifying imaging-derived MCI subgroups. The framework integrates gray-matter morphological features derived from structural magnetic resonance imaging and white-matter microstructural features derived from diffusion MRI. A Dual-Condition Variational Autoencoder was developed to separate biological variation from site-related effects and learn a site-robust normative representation using cognitively normal participants. Tract-specific MMSI scores were subsequently derived to quantify each participant's structural deviation from the normative reference. In a multi-center clinical cohort (N = 860), Gaussian mixture modeling of the tract-wise MMSI profiles identified lower- and higher-deviation MCI subgroups that exhibited significant differences in global cognition, delayed memory, and word recognition. Comparisons with conventional imaging biomarkers, machine-learning classifiers, and statistical harmonization methods further supported the value of the proposed framework. Evaluation in the Alzheimer's Disease Neuroimaging Initiative cohort demonstrated a concordance accuracy of 0.769 between the imaging-derived groups and the clinically predefined early- and late-MCI categories, exceeding the gray matter-only and white matter-only models by 7.7 and 12.3 percentage points, respectively. Transcriptomic association analysis identified KRT77 and CACNA1B as significantly associated with tract-specific MMSI scores after false discovery rate correction. Functional enrichment of the PLS-derived MMSI-associated gene set implicated biological processes related to neural signal transduction, learning and memory, calcium signaling, and immune responses. Together, these findings indicate that the MMSI framework provides a site-robust and interpretable approach for characterizing cross-sectional structural-deviation heterogeneity in MCI and offers preliminary biological support for the identified imaging patterns.