View-aligned hypergraph learning for Alzheimer's disease diagnosis with incomplete multi-modality data.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Data Analysis

Background:

• Alzheimer's disease (AD) diagnosis and its prodrome, mild cognitive impairment (MCI), often rely on multi-modality data.
• Incomplete datasets are common, posing challenges for traditional diagnostic models.
• Existing multi-view learning methods may overlook crucial inter-view coherence, limiting diagnostic performance.

Purpose of the Study:

• To develop an advanced machine learning framework for AD/MCI diagnosis using incomplete multi-modality data.
• To explicitly model and leverage the coherence among different data views (modalities).
• To enhance diagnostic accuracy by integrating information across heterogeneous data sources.

Main Methods:

• Proposed a view-aligned hypergraph learning (VAHL) method.
• Constructed view-specific hypergraphs using sparse representation from multi-modal data (MRI, PET, CSF).
• Developed a view-aligned hypergraph classification (VAHC) model with a regularizer to capture inter-view coherence and employed multi-view label fusion for final classification.

Main Results:

• The VAHL method demonstrated superior performance compared to state-of-the-art approaches on the ADNI-1 dataset.
• Effective utilization of incomplete multi-modality data for AD/MCI diagnosis was achieved.
• Explicitly modeling view coherence led to improved classification accuracy.

Conclusions:

• The proposed view-aligned hypergraph learning method offers a robust solution for AD/MCI diagnosis with incomplete multi-modality data.
• Modeling inter-view coherence is critical for optimizing diagnostic performance in complex neurodegenerative diseases.
• This approach holds promise for advancing early detection and diagnosis of Alzheimer's disease.