Linearized and Kernelized Sparse Multitask Learning for Predicting Cognitive Outcomes in Alzheimer's Disease.

Area of Science:

• Neuroimaging
• Machine Learning
• Biostatistics

Background:

• Alzheimer's disease (AD) poses significant financial and emotional burdens.
• Predicting cognitive performance from MRI and identifying biomarkers are crucial for AD research.
• Existing multitask learning (MTL) methods often assume linear relationships between MRI features and cognitive outcomes.

Purpose of the Study:

• To evaluate the performance of linear and nonlinear MTL methods for Alzheimer's disease cognitive score prediction.
• To develop an improved nonlinear MTL method for enhanced prediction and biomarker identification.
• To extend existing sparsity-inducing norms for broader applicability in neuroimaging analysis.

Main Methods:

• Utilized magnetic resonance imaging (MRI) data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.
• Applied and compared linear and nonlinear multitask learning (MTL) approaches, including multikernel MTL (MKMTL).
• Extended the ℓ2,1-norm to a generalized ℓq,1-norm (q ≥ 1) for feature selection.

Main Results:

• The proposed nonlinear ℓ2,1-ℓq-MKMTL method demonstrated superior cognitive score prediction performance compared to existing state-of-the-art methods.
• The method effectively fused multimodal data, enhancing its predictive power.
• Identified discriminative imaging biomarkers associated with cognitive decline in Alzheimer's disease.

Conclusions:

• Nonlinear MTL methods, particularly the proposed ℓ2,1-ℓq-MKMTL, offer significant advantages over linear approaches for predicting AD cognitive performance.
• The developed method provides a powerful tool for analyzing neuroimaging data and understanding Alzheimer's disease progression.
• This research contributes to advancing personalized medicine and early detection strategies for Alzheimer's disease.