Diagnosis of Alzheimer's disease using hypergraph p-Laplacian regularized multi-task feature learning

Area of Science:

• Neuroimaging
• Biomedical data analysis
• Machine learning for healthcare

Background:

• Multimodal data classification improves Alzheimer's disease (AD) diagnosis over single-modal methods.
• Existing methods often overlook non-linear, higher-order relationships within data, impacting model robustness.
• There is a need for advanced feature selection techniques that leverage complex data structures.

Purpose of the Study:

• To propose a novel hypergraph p-Laplacian regularized multi-task feature selection (HpMTFS) method for AD classification.
• To enhance the robustness and accuracy of AD diagnosis by incorporating higher-order data relationships.
• To jointly extract common features from multimodal data for improved classification performance.

Main Methods:

• HpMTFS method for feature selection across different data modalities (sMRI, FDG-PET, AV-45 PET).
• Utilizes group-sparsity regularization for joint feature extraction.
• Incorporates hypergraph p-Laplacian and Frobenius norm regularization terms to capture structural information and improve noise immunity.
• Multi-kernel support vector machine for multimodal feature fusion and classification.

Main Results:

• The HpMTFS method demonstrated superior performance compared to existing multimodal classification approaches.
• Experimental validation using data from 528 subjects in the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• The proposed method effectively utilizes higher-order structural information and enhances noise immunity.

Conclusions:

• The HpMTFS method offers a significant advancement in AD classification by effectively handling multimodal data.
• Capturing non-linear, higher-order relationships improves diagnostic model robustness and accuracy.
• This approach holds promise for more reliable early detection and diagnosis of Alzheimer's disease.