Atlas based sparse logistic regression for Alzheimer's Disease classification.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Sparse methods are crucial for high-dimensional neuroimaging data.
• Existing Group Lasso methods often group contiguous voxels, limiting biological relevance.
• Alzheimer's Disease (AD) and Mild Cognitive Impairment (MCI) classification benefits from effective feature selection.

Purpose of the Study:

• To develop novel sparse methods for AD and MCI classification.
• To introduce disease-related feature grouping strategies based on brain atlases.
• To improve classification performance and feature weight stability in neuroimaging.

Main Methods:

• Proposed two new feature grouping strategies using anatomical brain regions from a labeled atlas.
• Incorporated hierarchical grouping to account for bilateral symmetry in AD.
• Applied methods to MRI data from the Alzheimer's Disease Neuroimaging Initiative (ADNI).
• Compared performance and feature weight stability against existing sparse methods.

Main Results:

• The proposed methods achieved classification performance comparable to or better than existing approaches.
• Generated significantly more stable feature weights compared to conventional methods.
• Demonstrated the utility of anatomically and hierarchically defined feature groups.

Conclusions:

• Novel sparse grouping strategies based on brain atlases enhance classification of AD and MCI.
• The proposed methods offer improved stability in feature selection for neuroimaging studies.
• These findings have implications for developing more reliable diagnostic tools for neurodegenerative diseases.