Multiscale Functional Connectome Abnormality Predicts Cognitive Outcomes in Subcortical Ischemic Vascular Disease

Mianxin Liu,Yao Wang,Han Zhang,Qing Yang,Feng Shi,Yan Zhou,Dinggang Shen

DOI: https://doi.org/10.1093/cercor/bhab507

IF: 4.861

2022-02-05

Cerebral Cortex

Abstract:Abstract Subcortical ischemic vascular disease could induce subcortical vascular cognitive impairments (SVCIs), such as amnestic mild cognitive impairment (aMCI) and non-amnestic MCI (naMCI), or sometimes no cognitive impairment (NCI). Previous SVCI studies focused on focal structural lesions such as lacunes and microbleeds, while the functional connectivity networks (FCNs) from functional magnetic resonance imaging are drawing increasing attentions. Considering remarkable variations in structural lesion sizes, we expect that seeking abnormalities in the multiscale hierarchy of brain FCNs could be more informative to differentiate SVCI patients with varied outcomes (NCI, aMCI, and naMCI). Driven by this hypothesis, we first build FCNs based on the atlases at multiple spatial scales for group comparisons and found distributed FCN differences across different spatial scales. We then verify that combining multiscale features in a prediction model could improve differentiation accuracy among NCI, aMCI, and naMCI. Furthermore, we propose a graph convolutional network to integrate the naturally emerged multiscale features based on the brain network hierarchy, which significantly outperforms all other competing methods. In addition, the predictive features derived from our method consistently emphasize the limbic network in identifying aMCI across the different scales. The proposed analysis provides a better understanding of SVCI and may benefit its clinical diagnosis.

neurosciences