Mild Cognitive Impairment Identification Based on Multi-View Graph Convolutional Networks

Abstract

Though reversing the pathology of Alzheimer's disease (AD) has so far not been possible, a more tractable goal may be the prevention or slowing of the disease in its earliest stage, such as mild cognitive impairment (MCI). However, it is still challenge to identify patients with MCI in clinical practice. To address this challenge, we propose a new MCI identification method based on multi-view graph convolutional networks (MVGCNs). Firstly, we extract multiple morphological features based on multi-atlas from the imaging data of each subject. Then, we construct multiple population graphs (PGs) based on all experimental subjects using morphological features and non-imaging data of each subject. Afterwards, to obtain more discriminant features for MCI identification, multi-view graph convolutional networks are proposed for PGs based on each atlas. Finally, a new ensemble learning method is proposed to perform MCI identification task. Our proposed method is evaluated on 1449 subjects (including 301 subjects with AD, 779 subjects with MCI and 369 subjects with cognitively normal (CN)) from Alzheimer's Disease Neuroimaging Initiative (ADNI). Experimental results show that our proposed method achieves an accuracy of 90.8% for MCI/CN classification, and an accuracy of 88.6% for MCI/AD classification, respectively. Overall, our proposed method is effective and promising for automatic diagnosis of MCI in clinical practice.