Multi-scale Enhanced Graph Convolutional Network for Early Mild Cognitive Impairment Detection

Early mild cognitive impairment (EMCI) is an early stage of MCI, which can be detected by brain connectivity networks. To detect EMCI, we design a novel framework based on multi-scale enhanced GCN (MSE-GCN) in this paper, which fuses the functional and structural information from the resting-state functional magnetic resonance imaging and diffusion tensor imaging, respectively. Then both functional and structural information in connectivity networks are integrated via the local weighted clustering coefficients (LWCC), which are concatenated as the feature vectors to represent the vertices of population graph. Simultaneously, the subject’s gender and age in-formation is combined with the multi-modal neuroimaging feature to build a sparse graph. Then, we design multiple parallel GCN layers with different inputs by random walk embedding, which can identify the intrinsic MCI graph information from the embedding in GCN. Finally, we concatenate the output of all the GCN layers in the full connection layer for detection. The proposed method is capable of simultaneously representing the individual features and information associations among subjects from potential patients. The experimental results on the public Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset demonstrate that our proposed method achieves impressive EMCI identification performance compared with all competing methods.