Classification Study of Alzheimer’s Disease Based on Self-Attention Mechanism and DTI Imaging Using GCN

Abstract

Alzheimer's disease (AD) is a neurodegenerative disorder. Diffusion tensor imaging (DTI) provides information about the integrity of white matter fiber bundles that are related to the neuropathological mechanisms, and it is one of the commonly used techniques in AD research. In this study, we first divided each subject's brain into 90 regions based on the automated anatomical labeling (AAL) brain atlas. The average fractional anisotropy (FA) values between each pair of regions were applied to construct a brain network. We utilized the number of voxels with fibers passing through each brain region as the node features. The brain networks and node features were input into a novel graph convolutional neural network (GCN) structure involving the self-attention pooling mechanism proposed in this study to classify AD and normal controls (NC). The classification performance was compared among different preprocessed brain networks and node features. The final classification result achieved an accuracy of 87.5%. INDEX TERMSAlzheimer's disease; Diffusion tensor imaging; Image classification; Brain networks; Graph convolutional neural networks.