Modularity-Guided Functional Brain Network Analysis for Early-Stage Dementia Identification.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Functional brain network (FBN) analysis is crucial for detecting Alzheimer's disease (AD) and mild cognitive impairment (MCI).
• Existing methods often overlook the topological structure of FBNs, limiting classification performance and mechanistic understanding.
• Identifying discriminative and interpretable FBN features is essential for advancing AD/MCI diagnosis and research.

Purpose of the Study:

• To develop and evaluate a novel feature selection framework, modular-LASSO feature selection (MLFS), for automated AD/MCI classification.
• To explicitly incorporate modularity information from FBNs into the feature selection process.
• To improve the accuracy of identifying subjects with AD/MCI and predicting MCI conversion.

Main Methods:

• The proposed MLFS framework integrates signed spectral clustering to identify FBN modular structures.
• A modularity-induced group LASSO method is employed for discriminative feature selection.
• Classification is performed using a support vector machine (SVM) on selected features.

Main Results:

• The MLFS method demonstrated superior performance in identifying AD/MCI subjects compared to existing techniques.
• The framework effectively predicted the future conversion of MCI subjects to AD.
• Experiments on 563 resting-state fMRI scans from the ADNI database validated the method's efficacy.

Conclusions:

• The MLFS framework offers a powerful approach for analyzing FBNs in the context of AD and MCI.
• Explicitly modeling network modularity enhances the identification of discriminative and interpretable neuroimaging biomarkers.
• The method aids in understanding the pathological mechanisms underlying AD-related cognitive disorders.