Hypergraph based multi-task feature selection for multimodal classification of Alzheimer's disease.

Area of Science:

• Neuroimaging
• Machine Learning
• Biomedical Data Analysis

Background:

• Single-modality methods for Alzheimer's disease (AD) and mild cognitive impairment (MCI) diagnosis are less effective than multi-modality approaches.
• Existing multi-modality methods often overlook complex, high-order data structures, focusing only on pairwise relationships.

Purpose of the Study:

• To propose a novel hypergraph-based multi-task feature selection method for enhanced AD/MCI classification.
• To leverage high-order data structures for improved diagnostic accuracy in neurodegenerative diseases.

Main Methods:

• Feature selection was performed per modality using a group-sparsity regularizer for common feature identification.
• A hypergraph regularization term was integrated into multi-task feature selection to model complex subject relationships.
• A multi-kernel support vector machine fused selected features for final classification.

Main Results:

• The proposed method demonstrated superior classification performance compared to state-of-the-art multi-modality techniques.
• Experimental results on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset validated the method's effectiveness.
• Incorporating high-order structures significantly enhanced diagnostic accuracy for AD and MCI.

Conclusions:

• The hypergraph-based multi-task feature selection method effectively utilizes complex data structures for AD/MCI classification.
• This approach offers a significant advancement over traditional multi-modality methods by modeling intricate relationships.
• The findings suggest a promising direction for developing more accurate diagnostic tools for neurodegenerative diseases.