Attention-Based Framework for Alzheimer's Classification Using rs-fMRI
A new deep learning framework uses attention mechanisms to classify Alzheimer's disease directly from resting-state functional MRI connectivity matrices. By treating brain regions as tokens and applying Transformer-inspired self-attention, the model captures long-range functional dependencies across distributed brain networks without manual feature engineering. The approach addresses challenges of high dimensionality, noise, and complex inter-regional connectivity in rs-fMRI data. The framework was evaluated on a longitudinal cohort from the Alzheimer's Disease Neuroimaging Initiative (ADNI).
Key facts
- Attention-based deep learning framework for Alzheimer's disease classification
- Operates directly on rs-fMRI functional connectivity matrices
- Treats brain regions as tokens
- Uses Transformer-inspired self-attention mechanism
- Models long-range and global functional dependencies
- No reliance on manual feature engineering
- Evaluated on longitudinal cohort from ADNI
- Addresses high dimensionality, noise, and complex inter-regional dependencies
Entities
Institutions
- Alzheimer's Disease Neuroimaging Initiative (ADNI)