Diagnosis of Alzheimer's Disease Using View-Aligned Hypergraph Learning with Incomplete Multi-modality Data

Area of Science:

• Biomedical data analysis
• Machine learning for healthcare
• Neuroscience imaging

Background:

• Alzheimer's disease (AD) diagnosis often relies on incomplete multi-modal data.
• Existing multi-view learning methods may overlook crucial inter-view coherence, impacting diagnostic performance.
• Developing robust methods for AD diagnosis with incomplete data is critical.

Purpose of the Study:

• To propose a novel View-Aligned Hypergraph Learning (VAHL) method for Alzheimer's disease diagnosis.
• To explicitly model the coherence among different data modalities (views).
• To improve classification accuracy in diagnosing AD using incomplete multi-modal datasets.

Main Methods:

• Data divided into views based on modality combinations.
• Sparse representation-based hypergraph construction within each view.
• A View-Aligned Hypergraph Classification (VAHC) model incorporating a view-aligned regularizer.
• Multi-view label fusion for final classification.

Main Results:

• The proposed VAHL method demonstrated improved classification accuracy for AD/MCI diagnosis.
• Achieved at least a 4.6% increase in classification accuracy compared to state-of-the-art methods.
• Evaluated on the ADNI-1 database using MRI, PET, and CSF modalities from 807 subjects.

Conclusions:

• The VAHL method effectively utilizes incomplete multi-modal data for AD diagnosis.
• Explicitly modeling view coherence enhances the performance of multi-view learning in this context.
• The proposed approach offers a promising advancement for accurate Alzheimer's disease diagnosis.