Measuring the agreement between brain connectivity networks.

Area of Science:

• Neuroscience
• Machine Learning
• Network Science

Background:

• Comparing brain network structures is crucial for understanding brain function and disorders.
• Existing methods for assessing network similarity may lack sensitivity or dynamic range.
• Machine learning association measures offer a novel approach to quantify network similarity.

Purpose of the Study:

• To evaluate the efficacy of machine learning association measures for quantifying similarity between brain connectivity networks.
• To compare the performance of accuracy, Cohen's Kappa (K), and Area Under Curve (AUC) in assessing network similarity.
• To validate these measures in both simulated and real-world neuroimaging data.

Main Methods:

• Utilized accuracy, Cohen's Kappa (K), and Area Under Curve (AUC) as measures of association.
• Designed simulation studies with manipulated network similarity levels.
• Applied measures to surrogate data for methodological comparison and to real subject data for application assessment.

Main Results:

• All three measures (accuracy, K, AUC) successfully modulated their values based on network similarity.
• Area Under Curve (AUC) exhibited superior performance, demonstrating high dynamic range and sensitivity.
• AUC effectively synthesized the similarity between brain connectivity networks.

Conclusions:

• Machine learning association measures, particularly AUC, are effective tools for quantifying brain network similarity.
• These methods are applicable in both methodological studies comparing estimated to ground-truth networks and in clinical applications comparing subject networks to references.
• AUC's performance suggests its utility for sensitive and dynamic assessment of brain network structure similarity.