Identifying brain hierarchical structures associated with Alzheimer's disease using a regularized regression method...

Area of Science:

• Neuroimaging
• Statistical Modeling
• Alzheimer's Disease Research

Background:

• Brain segmentation is often modeled as hierarchical trees.
• Brain regional atrophy shows a marginal association with Alzheimer's disease (AD) outcomes.
• Existing methods may not fully capture the hierarchical nature of brain structure in AD studies.

Purpose of the Study:

• To propose a novel ℓ1-type regularization method for predictors with a hierarchical tree structure.
• To interpret model parameters using a path analysis perspective within a directed acyclic graph framework.
• To investigate the association between brain regional atrophy and Alzheimer's disease outcomes, specifically memory decline.

Main Methods:

• Developed an ℓ1-type regularization penalty for hierarchical tree-structured predictors.
• Interpreted model parameters through path analysis, regulating the total effect of predictors.
• Applied the method to brain structural MRI data from the Alzheimer's Disease Neuroimaging Initiative (ADNI).

Main Results:

• The proposed regularization method ensures consistency of the coefficient estimator in ℓ2-norm and model selection consistency.
• The approach successfully identified brain regions where atrophy correlates with memory decline in AD patients.
• Findings indicate that the impact of atrophy on memory deficits is localized to specific brain regions across various segmentation levels.

Conclusions:

• The proposed regularization technique is effective for analyzing hierarchical data in neuroimaging studies, particularly for Alzheimer's disease.
• This method provides insights into the localized effects of brain atrophy on cognitive decline, specifically memory.
• The study highlights the importance of considering hierarchical structures in brain segmentation for understanding AD progression.