Detecting functional connectivity in the resting brain: a comparison between ICA and CCA

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Biophysics

Background:

• Resting-state functional connectivity (rsFC) is crucial for understanding brain function.
• Independent Component Analysis (ICA) and Cross-Correlation Analysis (CCA) are common rsFC detection methods.
• Evaluating and comparing these methods is essential for accurate brain network analysis.

Purpose of the Study:

• To jointly evaluate Independent Component Analysis (ICA) and Cross-Correlation Analysis (CCA) for resting-state functional connectivity.
• To investigate the impact of parameter choices (number of components for ICA, seed selection for CCA) on analysis outcomes.
• To compare the performance and robustness of ICA versus CCA.

Main Methods:

• Simulated data and in vivo functional magnetic resonance imaging (fMRI) data from 10 healthy subjects were used.
• The influence of the number of independent components was systematically studied for ICA.
• The effect of seed selection criteria was examined for CCA.

Main Results:

• Significant differences were observed between ICA and CCA.
• ICA demonstrated superior performance compared to CCA and showed robustness to structured noise.
• ICA performance was sensitive to an insufficient number of components.
• Converting ICA maps to z-maps for thresholding led to a slight overestimation of false positives, though generally acceptable.

Conclusions:

• ICA is a more reliable method for detecting resting-state functional connectivity than CCA.
• Parameter selection, particularly the number of components in ICA, is critical for optimal results.
• CCA results are highly dependent on the chosen seed locations, limiting its generalizability.