Structured Sparse Kernel Learning for Imaging Genetics Based Alzheimer's Disease Diagnosis.

Area of Science:

• Neuroimaging
• Genetics
• Machine Learning

Background:

• Alzheimer's disease (AD) diagnosis requires integrating complex multimodal data.
• Existing methods struggle with feature redundancy and cross-modality fusion.
• Advanced machine learning techniques are needed for accurate AD prediction.

Purpose of the Study:

• To develop a novel kernel-learning framework for integrating multimodal imaging and genetic data for AD diagnosis.
• To introduce a structured sparsity regularizer for effective feature selection and fusion across modalities.
• To improve the accuracy of Alzheimer's disease diagnosis using a combination of brain imaging and genetic information.

Main Methods:

• A kernel-learning approach representing features as kernels, grouped by modality.
• A novel structured sparsity regularizer for simultaneous sparse feature selection within modalities and dense fusion across modalities.
• Evaluation using magnetic resonance imaging (MRI), positron emission tomography (PET), and single-nucleotide polymorphism (SNP) data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.

Main Results:

• The proposed method significantly improved prediction accuracy for Alzheimer's disease.
• The approach successfully identified relevant brain regions associated with AD.
• Key single-nucleotide polymorphisms (SNPs) implicated in Alzheimer's disease were discovered.

Conclusions:

• The kernel-learning based method effectively integrates multimodal data for enhanced AD diagnosis.
• The novel structured sparsity regularizer facilitates robust feature selection and fusion.
• This approach holds promise for improving early detection and understanding of Alzheimer's disease.