Single-index models with functional connectivity network predictors.

Area of Science:

• Neuroimaging
• Brain Connectivity Analysis
• Statistical Modeling

Background:

• Functional connectivity, derived from functional magnetic resonance imaging (fMRI) time series, is increasingly used for clinical outcome prediction.
• Existing methods may not fully capture complex, nonlinear relationships within brain networks.

Purpose of the Study:

• To propose a novel single-index model for analyzing sparse functional connectivity networks.
• To accommodate potentially nonlinear relationships between brain connectivity and response variables.
• To identify specific brain region interactions associated with outcomes.

Main Methods:

• Developed a single-index model linking response variables with sparse functional connectivity network predictors.
• Employed unspecified smooth functions to model nonlinear relationships.
• Utilized sparsity constraints to identify significant network associations and regional contributions.

Main Results:

• The proposed model effectively identified associations between brain region interactions and response variables in simulation studies.
• Application to Human Connectome Project resting-state fMRI data successfully modeled fluid intelligence and sex.
• Predictive links between specific brain regions and these attributes were identified.

Conclusions:

• The single-index model offers a powerful approach for analyzing functional connectivity data.
• The method enhances the ability to identify meaningful predictive relationships within brain networks.
• This approach has significant potential for advancing neuroimaging-based prediction and classification.