Individual-specific fMRI-Subspaces improve functional connectivity prediction of behavior.

Area of Science:

• Neuroscience
• Cognitive Science
• Psychology

Background:

• Resting-state functional connectivity (RSFC) is of significant interest for predicting human behavior.
• Effective behavioral prediction theoretically requires distinct RSFC patterns among individuals.
• Common signals shared across participants might obscure subject-specific patterns, hindering prediction accuracy.

Purpose of the Study:

• To investigate whether removing common resting-state functional magnetic resonance imaging (rs-fMRI) signals shared across participants improves behavioral prediction.
• To identify and characterize common signals within rs-fMRI data using the COBE technique.
• To assess the impact of removing these common signals on the prediction of diverse behavioral measures.

Main Methods:

• Utilized resting-state fMRI data from 803 participants in the Human Connectome Project (HCP).
• Applied the Common and Orthogonal Basis Extraction (COBE) technique to decompose rs-fMRI runs into common (group-level) and subject-specific subspaces.
• Removed identified common COBE components from the data before assessing behavioral prediction.

Main Results:

• The first common COBE component of the first HCP run localized to the visual cortex, appearing run-specific.
• Subsequent common COBE components (second from the first run, first from remaining runs) were similar and localized to the default mode network.
• Removing these identified common components improved behavioral prediction by an average of 11.7% across 58 behavioral measures.

Conclusions:

• Common signals in rs-fMRI data, potentially representing run-specific or state-specific effects, can be identified and removed.
• Removing these shared signals enhances the subject-specific information within RSFC.
• This approach significantly improves the prediction of human behavior across cognitive, emotional, and personality domains.