Functional Brain Networks: Does the Choice of Dependency Estimator and Binarization Method Matter?

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging Analysis

Background:

• Brain function is often modeled as a complex network, with regions as nodes and connections as edges.
• Functional brain networks are typically derived from connectivity matrices, which require binarization to reduce noise.
• Existing literature presents varied findings on network properties in Alzheimer's Disease (AD), potentially due to methodological differences.

Purpose of the Study:

• To investigate the topological properties of electroencephalography (EEG)-based functional brain networks in Alzheimer's Disease (AD).
• To evaluate the impact of different dependency estimation and network binarization methods on AD-related network abnormalities.
• To identify methods that best distinguish between AD patients and healthy controls in functional brain network analysis.

Main Methods:

• Connectivity estimation using Pearson correlation, coherence, phase order parameter, and synchronization likelihood.
• Binarization of weighted connectivity matrices using Minimum Spanning Tree (MST), Minimum Connected Component (MCC), uniform threshold, and density-preserving methods.
• Analysis of topological properties of constructed EEG-based functional networks.

Main Results:

• The identification of AD-related abnormalities is highly sensitive to the chosen methods for dependency estimation and binarization.
• Functional networks constructed using the coherence method for connectivity estimation and MCC for binarization demonstrated the most significant differences between AD and healthy subjects.
• Other combinations of methods yielded less pronounced or non-significant differences, highlighting the importance of methodological selection.

Conclusions:

• Methodological choices in functional brain network analysis critically influence the detection of Alzheimer's Disease (AD) characteristics.
• The coherence and MCC binarization approach appears promising for identifying robust network differences in AD.
• Standardizing or carefully considering analysis methods is crucial for interpreting and comparing results in AD neuroimaging research.