Classification of mild cognitive impairment developmental trajectories using multispatial scale structural brain networks.

Abstract

Objective: In this study, we investigated the interactions between brain regions at different scales in patients with mild cognitive impairment (MCI) to classify patients with MCI who may develop Alzheimer's disease (MCI-Converter (MCI-c)) and those with stable cognitive states (MCI-Stable (MCI-s)) at multiple spatial scales.

Methods: We divided the brain into 210, 40, and 12 regions, respectively, based on anatomical a priori, and then extracted six morphological features. Based on this, an intralayer structural brain network was constructed to detect connections between brain regions at different levels, and an interlayer network was constructed to explore connections between different spatial scales. Then these two networks were merged into a whole-brain network and trained the classifier after feature selection.

Results: Our study successfully identified meaningful connectivity features for precise classification, achieving an accuracy of 92.41%. In addition, some frequently reported abnormal brain regions were localized to more precise regions.

Conclusion: The human brain is a complex system with multiple spatial and temporal scales and multiple levels, showing a large number of emergent phenomena. Understanding the hierarchical relationship between brain structure and function is crucial. The network we constructed is not only important for MCI classification, but also holds promise for investigating other neurological conditions and elucidating brain development processes. Limitations include that model training and evaluation used only the Alzheimer's Disease Neuroimaging Initiative cohort; independent cohort validation is required to confirm generalizability. Moreover, integration with other imaging modalities (e.g. functional MRI and PET) may further improve prediction and will be explored in future work.