Snowball ICA

Snowball ICA implements a model-order-free independent component analysis (ICA) strategy to extract spatial maps of signal sources from functional magnetic resonance imaging (fMRI) data without requiring a pre-specified model order.

Key Features:

• Model-order-free ICA: Operates without a pre-determined model order for component extraction from fMRI data.
• Network-scale agnostic collection: Systematically collects relevant information for each brain network across scales from the fMRI data.
• Improved component estimation: Produces more accurate spatial maps of signal sources compared to traditional ICA methods affected by model order choice.
• Robustness to noise-related mutual information: Addresses increased mutual information between widely distributed brain networks and noise that degrades high model-order ICA performance.
• Validation framework: Performance has been evaluated using simulation studies and analyses of in vivo resting-state fMRI data.
• Focus on spatial map accuracy: Targets accurate estimation of spatial maps of signal sources derived from fMRI.

Scientific Applications:

• Brain network analysis: Analysis of widely distributed brain networks in fMRI data without selecting a model order.
• Resting-state fMRI analysis: Extraction and assessment of independent components from in vivo resting-state fMRI datasets.
• Spatial source mapping: Reconstruction of spatial maps of signal sources for neuroimaging studies.
• Simulation-based benchmarking: Use in simulation studies to assess and compare ICA accuracy under varying conditions.

Methodology:

Applies independent component analysis (ICA) without a pre-specified model order by systematically collecting network-specific information from fMRI data, and validates performance with simulation studies and in vivo resting-state fMRI analyses.