Gray Matter-Guided Attention Network for AD Diagnosis Using Structural MRI

In recent years, deep learning-based methods using structural magnetic resonance imaging (sMRI) have shown promising results in diagnosing cognitive impairment disorders. However, many existing methods do not fully leverage the characteristics of brain images. In this study, we constructed an end-to-end network with a gray matter-guided attention mechanism for Alzheimer’s disease (AD) diagnosis. Gray matter atrophy in the brain is a sensitive imaging marker for AD. The network can achieve higher diagnostic performance by incorporating gray matter information and focusing on related key regions in the brain when encoding brain image. In addition, we use the 3D Grad CAM applied to the proposed model to show the area of interest of the network in the diagnostic tasks and a certain interpretability. We evaluated our method on the ADNI dataset in AD diagnostic tasks and demonstrated that it is more effective compared to several methods.