Connectivity Regression

Area of Science:

• Neuroscience
• Biostatistics
• Machine Learning

Background:

• Brain functional connectivity networks vary across individuals, impacting healthy aging and disease.
• Understanding these variations is crucial for neuroscience and clinical applications.

Purpose of the Study:

• Introduce Connectivity Regression (ConnReg), a novel framework for analyzing subject-specific functional connectivity networks.
• Account for within-network inter-edge dependence to improve regression analysis.

Main Methods:

• ConnReg uses a multivariate Fisher's transformation for network data projection.
• Employs penalized multivariate regression to induce sparsity in coefficients and covariance.
• Utilizes permutation tests for multiplicity-adjusted inference and stability selection for edge identification.

Main Results:

• Simulation studies validate ConnReg's inferential properties and efficiency.
• Accounting for within-network inter-edge dependence enhances estimation, inference power, and selection accuracy.
• ConnReg application to Human Connectome Project data reveals insights into connectivity variations.

Conclusions:

• ConnReg provides a robust framework for analyzing functional connectivity data.
• The method improves understanding of how covariates like language processing and brain structure relate to connectivity.
• This approach has implications for studying brain aging, neurological disorders, and individual differences.