Joint selection of brain network nodes and edges for MCI identification

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Functional brain graphs (FBGs) represent brain interactions from fMRI data, crucial for identifying mild cognitive impairment (MCI).
• Feature selection from FBGs is vital for reducing computational load, preventing overfitting, and discovering biomarkers.
• Current methods often focus on either node- or edge-based features, potentially missing crucial information.

Purpose of the Study:

• To propose a novel method for jointly selecting nodes and edges from functional brain graphs.
• To enhance the classification performance and interpretability of MCI identification.
• To identify discriminative brain regions and their connections for potential biomarker discovery.

Main Methods:

• A novel approach is presented to jointly select nodes and edges from functional brain graphs (FBGs).
• Edges are assigned to node groups, followed by the application of sparse group least absolute shrinkage and selection operator (sgLASSO).
• This technique simultaneously identifies discriminative brain regions and their interconnections.

Main Results:

• The proposed method demonstrated superior classification performance compared to existing state-of-the-art techniques.
• Analysis of the selected brain network features led to the discovery of potential biomarkers for MCI diagnosis.
• The joint node and edge selection approach enhances the interpretability of classification results.

Conclusions:

• A novel method for joint node and edge selection in functional brain graphs (FBGs) has been successfully developed.
• This approach offers improved accuracy and interpretability for identifying mild cognitive impairment (MCI).
• The findings highlight the potential for discovering new biomarkers for early Alzheimer's disease detection.