The integration of AI, big data, and magnetic resonance imaging (MRI) has significantly advanced healthcare, particularly in the early diagnosis and treatment of neurodegenerative diseases. Alzheimer’s Disease (AD), which predominantly affects the elderly, and its precursor, Mild Cognitive Impairment (MCI), present considerable challenges in early detection due to the complex structural changes in the brain. Traditional diagnostic methods often struggle to capture these intricate changes. To address this, the systematic Brain Informatics methodology is reconsidered to realize evidence combination based on dual-structural representation and fusion computing via attention-based graph. In particular, the DS-AGF (Dual-Structural Representation Learning with Attention-Based Graph Fusion) model was proposed for hierarchical MCI-AD diagnosis. This model employs a dense convolution unit and a node-optimized convolution unit to learn multi-level, fine-grained brain representations. Additionally, an attention-based graph representation fusion unit is introduced, enabling the integration of these representations and allowing the model to capture both local and global relationships between brain regions. This approach enhances the network feature learning ability on critical features, enabling multi-view visualization of key brain regions and their connectivity.