Dual feature correlation guided multi-task learning for Alzheimer's disease prediction.

Area of Science:

• Neuroimaging and Computational Neuroscience
• Biomedical Data Analysis

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder impacting cognitive functions.
• Predicting cognitive decline and identifying neuroimaging biomarkers are crucial for AD research.
• Existing machine learning models face the curse of dimensionality, and Multi-Task Feature Learning (MTFL) overlooks correlations within imaging data.

Purpose of the Study:

• To develop an advanced framework for predicting cognitive scores from neuroimaging measures in Alzheimer's disease.
• To identify stable and sensitive neuroimaging biomarkers for AD.
• To address the limitations of existing models by incorporating both task and feature correlation structures.

Main Methods:

• Proposed a generalized multi-task learning framework integrating task and feature correlation structures.
• Introduced a novel feature-aware sparsity-inducing norm (FAS-norm) penalty to leverage correlations among brain imaging features.
• Developed an optimization algorithm using the alternating direction method of multipliers (ADMM) for non-smooth problems.

Main Results:

• The proposed models demonstrated improved prediction accuracy compared to standard MTFL.
• Achieved an average 4.28% decrease in cross-sectional analysis error and a 7.97% decrease in longitudinal ADAS-Cog score prediction.
• Identified key biomarkers including the hippocampus, lateral ventricle, and corpus callosum.

Conclusions:

• The novel multi-task learning framework effectively predicts cognitive scores and identifies robust biomarkers in Alzheimer's disease.
• Incorporating feature correlations via FAS-norm enhances prediction accuracy and biomarker stability.
• The findings offer valuable tools for clinical diagnosis and understanding of Alzheimer's disease progression.