Predicting behavior through dynamic modes in resting-state fMRI data.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Imaging

Background:

• Resting-state functional connectivity (FC) dynamics offer insights into brain-behavior relationships.
• Dynamic mode decomposition (DMD) is a method for characterizing FC dynamics.
• The predictive power of DMD-derived dynamic modes (DMs) for individual behavior is not well understood.

Purpose of the Study:

• To develop and validate a method for predicting individual behavioral differences using DMs from rs-fMRI data.
• To investigate the contribution of DMs across different frequency bands to behavioral prediction.
• To compare the efficacy of DMD with conventional methods like independent component analysis (ICA).

Main Methods:

• Computed subject-specific DMs from rs-fMRI data.
• Utilized multivariate pattern analysis on a Gram matrix of DMs to predict 59 behavioral measures.
• Employed permutation testing for statistical validation.
• Analyzed the contribution of DMs within specific frequency bands (0-0.1 to 0.6-0.7 Hz).

Main Results:

• DMD successfully predicted individual behavior, outperforming spatial and temporal ICA.
• Cognitive measures were most frequently predicted with significant accuracy.
• DMs in lower frequency bands (<0.2 Hz) were primary contributors to prediction.
• The spatial structures of predictive DMs resembled known resting-state networks.

Conclusions:

• DMD is an efficient method for extracting predictive spatiotemporal features from rs-fMRI data.
• DMD-derived DMs provide valuable information for understanding individual differences in behavior, particularly cognition.
• Lower frequency bands (<0.2 Hz) contain crucial information for linking brain connectivity dynamics to behavior.